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 <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <asm/uaccess.h>
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
75 extern union ib_gid zgid;
78 /* If link layer is Ethernet, this is RoCE V1 */
81 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
85 #define ROCE_V2_UDP_DPORT 4791
87 enum ib_gid_type gid_type;
88 struct net_device *ndev;
92 /* IB values map to NodeInfo:NodeType. */
102 /* set the local administered indication */
103 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
106 enum rdma_transport_type {
108 RDMA_TRANSPORT_IWARP,
109 RDMA_TRANSPORT_USNIC,
110 RDMA_TRANSPORT_USNIC_UDP
113 enum rdma_protocol_type {
117 RDMA_PROTOCOL_USNIC_UDP
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
123 enum rdma_network_type {
125 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
132 if (network_type == RDMA_NETWORK_IPV4 ||
133 network_type == RDMA_NETWORK_IPV6)
134 return IB_GID_TYPE_ROCE_UDP_ENCAP;
136 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 return IB_GID_TYPE_IB;
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
143 if (gid_type == IB_GID_TYPE_IB)
144 return RDMA_NETWORK_IB;
146 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 return RDMA_NETWORK_IPV4;
149 return RDMA_NETWORK_IPV6;
152 enum rdma_link_layer {
153 IB_LINK_LAYER_UNSPECIFIED,
154 IB_LINK_LAYER_INFINIBAND,
155 IB_LINK_LAYER_ETHERNET,
158 enum ib_device_cap_flags {
159 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
160 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
161 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
162 IB_DEVICE_RAW_MULTI = (1 << 3),
163 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
164 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
165 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
166 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
167 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
168 IB_DEVICE_INIT_TYPE = (1 << 9),
169 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
170 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
171 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
172 IB_DEVICE_SRQ_RESIZE = (1 << 13),
173 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
176 * This device supports a per-device lkey or stag that can be
177 * used without performing a memory registration for the local
178 * memory. Note that ULPs should never check this flag, but
179 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 * which will always contain a usable lkey.
182 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
183 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16),
184 IB_DEVICE_MEM_WINDOW = (1 << 17),
186 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 * messages and can verify the validity of checksum for
189 * incoming messages. Setting this flag implies that the
190 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
192 IB_DEVICE_UD_IP_CSUM = (1 << 18),
193 IB_DEVICE_UD_TSO = (1 << 19),
194 IB_DEVICE_XRC = (1 << 20),
197 * This device supports the IB "base memory management extension",
198 * which includes support for fast registrations (IB_WR_REG_MR,
199 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
200 * also be set by any iWarp device which must support FRs to comply
201 * to the iWarp verbs spec. iWarp devices also support the
202 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
205 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
206 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
207 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
208 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
209 IB_DEVICE_RC_IP_CSUM = (1 << 25),
210 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
212 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 * support execution of WQEs that involve synchronization
214 * of I/O operations with single completion queue managed
217 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
218 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
219 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
220 IB_DEVICE_ON_DEMAND_PAGING = (1 << 31),
221 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
222 IB_DEVICE_VIRTUAL_FUNCTION = ((u64)1 << 33),
225 enum ib_signature_prot_cap {
226 IB_PROT_T10DIF_TYPE_1 = 1,
227 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
228 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
231 enum ib_signature_guard_cap {
232 IB_GUARD_T10DIF_CRC = 1,
233 IB_GUARD_T10DIF_CSUM = 1 << 1,
242 enum ib_odp_general_cap_bits {
243 IB_ODP_SUPPORT = 1 << 0,
246 enum ib_odp_transport_cap_bits {
247 IB_ODP_SUPPORT_SEND = 1 << 0,
248 IB_ODP_SUPPORT_RECV = 1 << 1,
249 IB_ODP_SUPPORT_WRITE = 1 << 2,
250 IB_ODP_SUPPORT_READ = 1 << 3,
251 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
255 uint64_t general_caps;
257 uint32_t rc_odp_caps;
258 uint32_t uc_odp_caps;
259 uint32_t ud_odp_caps;
260 } per_transport_caps;
263 enum ib_cq_creation_flags {
264 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
265 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
268 struct ib_cq_init_attr {
274 struct ib_device_attr {
276 __be64 sys_image_guid;
284 u64 device_cap_flags;
294 int max_qp_init_rd_atom;
295 int max_ee_init_rd_atom;
296 enum ib_atomic_cap atomic_cap;
297 enum ib_atomic_cap masked_atomic_cap;
304 int max_mcast_qp_attach;
305 int max_total_mcast_qp_attach;
312 unsigned int max_fast_reg_page_list_len;
314 u8 local_ca_ack_delay;
317 struct ib_odp_caps odp_caps;
318 uint64_t timestamp_mask;
319 uint64_t hca_core_clock; /* in KHZ */
330 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
333 case IB_MTU_256: return 256;
334 case IB_MTU_512: return 512;
335 case IB_MTU_1024: return 1024;
336 case IB_MTU_2048: return 2048;
337 case IB_MTU_4096: return 4096;
348 IB_PORT_ACTIVE_DEFER = 5
351 enum ib_port_cap_flags {
353 IB_PORT_NOTICE_SUP = 1 << 2,
354 IB_PORT_TRAP_SUP = 1 << 3,
355 IB_PORT_OPT_IPD_SUP = 1 << 4,
356 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
357 IB_PORT_SL_MAP_SUP = 1 << 6,
358 IB_PORT_MKEY_NVRAM = 1 << 7,
359 IB_PORT_PKEY_NVRAM = 1 << 8,
360 IB_PORT_LED_INFO_SUP = 1 << 9,
361 IB_PORT_SM_DISABLED = 1 << 10,
362 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
363 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
364 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
365 IB_PORT_CM_SUP = 1 << 16,
366 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
367 IB_PORT_REINIT_SUP = 1 << 18,
368 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
369 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
370 IB_PORT_DR_NOTICE_SUP = 1 << 21,
371 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
372 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
373 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
374 IB_PORT_CLIENT_REG_SUP = 1 << 25,
375 IB_PORT_IP_BASED_GIDS = 1 << 26,
385 static inline int ib_width_enum_to_int(enum ib_port_width width)
388 case IB_WIDTH_1X: return 1;
389 case IB_WIDTH_4X: return 4;
390 case IB_WIDTH_8X: return 8;
391 case IB_WIDTH_12X: return 12;
405 struct ib_protocol_stats {
409 struct iw_protocol_stats {
412 u64 ipInTooBigErrors;
415 u64 ipInUnknownProtos;
416 u64 ipInTruncatedPkts;
419 u64 ipOutForwDatagrams;
451 union rdma_protocol_stats {
452 struct ib_protocol_stats ib;
453 struct iw_protocol_stats iw;
456 /* Define bits for the various functionality this port needs to be supported by
459 /* Management 0x00000FFF */
460 #define RDMA_CORE_CAP_IB_MAD 0x00000001
461 #define RDMA_CORE_CAP_IB_SMI 0x00000002
462 #define RDMA_CORE_CAP_IB_CM 0x00000004
463 #define RDMA_CORE_CAP_IW_CM 0x00000008
464 #define RDMA_CORE_CAP_IB_SA 0x00000010
465 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
467 /* Address format 0x000FF000 */
468 #define RDMA_CORE_CAP_AF_IB 0x00001000
469 #define RDMA_CORE_CAP_ETH_AH 0x00002000
471 /* Protocol 0xFFF00000 */
472 #define RDMA_CORE_CAP_PROT_IB 0x00100000
473 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
474 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
475 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
477 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
478 | RDMA_CORE_CAP_IB_MAD \
479 | RDMA_CORE_CAP_IB_SMI \
480 | RDMA_CORE_CAP_IB_CM \
481 | RDMA_CORE_CAP_IB_SA \
482 | RDMA_CORE_CAP_AF_IB)
483 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
484 | RDMA_CORE_CAP_IB_MAD \
485 | RDMA_CORE_CAP_IB_CM \
486 | RDMA_CORE_CAP_AF_IB \
487 | RDMA_CORE_CAP_ETH_AH)
488 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
489 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_CM \
492 | RDMA_CORE_CAP_AF_IB \
493 | RDMA_CORE_CAP_ETH_AH)
494 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
495 | RDMA_CORE_CAP_IW_CM)
496 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
497 | RDMA_CORE_CAP_OPA_MAD)
499 struct ib_port_attr {
501 enum ib_port_state state;
503 enum ib_mtu active_mtu;
523 enum ib_device_modify_flags {
524 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
525 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
528 struct ib_device_modify {
533 enum ib_port_modify_flags {
534 IB_PORT_SHUTDOWN = 1,
535 IB_PORT_INIT_TYPE = (1<<2),
536 IB_PORT_RESET_QKEY_CNTR = (1<<3)
539 struct ib_port_modify {
540 u32 set_port_cap_mask;
541 u32 clr_port_cap_mask;
549 IB_EVENT_QP_ACCESS_ERR,
553 IB_EVENT_PATH_MIG_ERR,
554 IB_EVENT_DEVICE_FATAL,
555 IB_EVENT_PORT_ACTIVE,
558 IB_EVENT_PKEY_CHANGE,
561 IB_EVENT_SRQ_LIMIT_REACHED,
562 IB_EVENT_QP_LAST_WQE_REACHED,
563 IB_EVENT_CLIENT_REREGISTER,
567 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
570 struct ib_device *device;
577 enum ib_event_type event;
580 struct ib_event_handler {
581 struct ib_device *device;
582 void (*handler)(struct ib_event_handler *, struct ib_event *);
583 struct list_head list;
586 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
588 (_ptr)->device = _device; \
589 (_ptr)->handler = _handler; \
590 INIT_LIST_HEAD(&(_ptr)->list); \
593 struct ib_global_route {
602 __be32 version_tclass_flow;
610 union rdma_network_hdr {
613 /* The IB spec states that if it's IPv4, the header
614 * is located in the last 20 bytes of the header.
617 struct iphdr roce4grh;
622 IB_MULTICAST_QPN = 0xffffff
625 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
626 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
633 IB_RATE_PORT_CURRENT = 0,
634 IB_RATE_2_5_GBPS = 2,
642 IB_RATE_120_GBPS = 10,
643 IB_RATE_14_GBPS = 11,
644 IB_RATE_56_GBPS = 12,
645 IB_RATE_112_GBPS = 13,
646 IB_RATE_168_GBPS = 14,
647 IB_RATE_25_GBPS = 15,
648 IB_RATE_100_GBPS = 16,
649 IB_RATE_200_GBPS = 17,
650 IB_RATE_300_GBPS = 18
654 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
655 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
656 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
657 * @rate: rate to convert.
659 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
662 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
663 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
664 * @rate: rate to convert.
666 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
670 * enum ib_mr_type - memory region type
671 * @IB_MR_TYPE_MEM_REG: memory region that is used for
672 * normal registration
673 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
674 * signature operations (data-integrity
676 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
677 * register any arbitrary sg lists (without
678 * the normal mr constraints - see
683 IB_MR_TYPE_SIGNATURE,
689 * IB_SIG_TYPE_NONE: Unprotected.
690 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
692 enum ib_signature_type {
698 * Signature T10-DIF block-guard types
699 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
700 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
702 enum ib_t10_dif_bg_type {
708 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
710 * @bg_type: T10-DIF block guard type (CRC|CSUM)
711 * @pi_interval: protection information interval.
712 * @bg: seed of guard computation.
713 * @app_tag: application tag of guard block
714 * @ref_tag: initial guard block reference tag.
715 * @ref_remap: Indicate wethear the reftag increments each block
716 * @app_escape: Indicate to skip block check if apptag=0xffff
717 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
718 * @apptag_check_mask: check bitmask of application tag.
720 struct ib_t10_dif_domain {
721 enum ib_t10_dif_bg_type bg_type;
729 u16 apptag_check_mask;
733 * struct ib_sig_domain - Parameters for signature domain
734 * @sig_type: specific signauture type
735 * @sig: union of all signature domain attributes that may
736 * be used to set domain layout.
738 struct ib_sig_domain {
739 enum ib_signature_type sig_type;
741 struct ib_t10_dif_domain dif;
746 * struct ib_sig_attrs - Parameters for signature handover operation
747 * @check_mask: bitmask for signature byte check (8 bytes)
748 * @mem: memory domain layout desciptor.
749 * @wire: wire domain layout desciptor.
751 struct ib_sig_attrs {
753 struct ib_sig_domain mem;
754 struct ib_sig_domain wire;
757 enum ib_sig_err_type {
764 * struct ib_sig_err - signature error descriptor
767 enum ib_sig_err_type err_type;
774 enum ib_mr_status_check {
775 IB_MR_CHECK_SIG_STATUS = 1,
779 * struct ib_mr_status - Memory region status container
781 * @fail_status: Bitmask of MR checks status. For each
782 * failed check a corresponding status bit is set.
783 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
786 struct ib_mr_status {
788 struct ib_sig_err sig_err;
792 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
794 * @mult: multiple to convert.
796 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
799 struct ib_global_route grh;
813 IB_WC_LOC_EEC_OP_ERR,
818 IB_WC_LOC_ACCESS_ERR,
819 IB_WC_REM_INV_REQ_ERR,
820 IB_WC_REM_ACCESS_ERR,
823 IB_WC_RNR_RETRY_EXC_ERR,
824 IB_WC_LOC_RDD_VIOL_ERR,
825 IB_WC_REM_INV_RD_REQ_ERR,
828 IB_WC_INV_EEC_STATE_ERR,
830 IB_WC_RESP_TIMEOUT_ERR,
834 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
845 IB_WC_MASKED_COMP_SWAP,
846 IB_WC_MASKED_FETCH_ADD,
848 * Set value of IB_WC_RECV so consumers can test if a completion is a
849 * receive by testing (opcode & IB_WC_RECV).
852 IB_WC_RECV_RDMA_WITH_IMM
857 IB_WC_WITH_IMM = (1<<1),
858 IB_WC_WITH_INVALIDATE = (1<<2),
859 IB_WC_IP_CSUM_OK = (1<<3),
860 IB_WC_WITH_SMAC = (1<<4),
861 IB_WC_WITH_VLAN = (1<<5),
862 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
868 struct ib_cqe *wr_cqe;
870 enum ib_wc_status status;
871 enum ib_wc_opcode opcode;
885 u8 port_num; /* valid only for DR SMPs on switches */
891 enum ib_cq_notify_flags {
892 IB_CQ_SOLICITED = 1 << 0,
893 IB_CQ_NEXT_COMP = 1 << 1,
894 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
895 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
903 enum ib_srq_attr_mask {
904 IB_SRQ_MAX_WR = 1 << 0,
905 IB_SRQ_LIMIT = 1 << 1,
914 struct ib_srq_init_attr {
915 void (*event_handler)(struct ib_event *, void *);
917 struct ib_srq_attr attr;
918 enum ib_srq_type srq_type;
922 struct ib_xrcd *xrcd;
936 * Maximum number of rdma_rw_ctx structures in flight at a time.
937 * ib_create_qp() will calculate the right amount of neededed WRs
938 * and MRs based on this.
950 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
951 * here (and in that order) since the MAD layer uses them as
952 * indices into a 2-entry table.
961 IB_QPT_RAW_ETHERTYPE,
962 IB_QPT_RAW_PACKET = 8,
966 /* Reserve a range for qp types internal to the low level driver.
967 * These qp types will not be visible at the IB core layer, so the
968 * IB_QPT_MAX usages should not be affected in the core layer
970 IB_QPT_RESERVED1 = 0x1000,
982 enum ib_qp_create_flags {
983 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
984 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
985 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
986 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
987 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
988 IB_QP_CREATE_NETIF_QP = 1 << 5,
989 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
990 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
991 /* reserve bits 26-31 for low level drivers' internal use */
992 IB_QP_CREATE_RESERVED_START = 1 << 26,
993 IB_QP_CREATE_RESERVED_END = 1 << 31,
997 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
998 * callback to destroy the passed in QP.
1001 struct ib_qp_init_attr {
1002 void (*event_handler)(struct ib_event *, void *);
1004 struct ib_cq *send_cq;
1005 struct ib_cq *recv_cq;
1007 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1008 struct ib_qp_cap cap;
1009 enum ib_sig_type sq_sig_type;
1010 enum ib_qp_type qp_type;
1011 enum ib_qp_create_flags create_flags;
1014 * Only needed for special QP types, or when using the RW API.
1019 struct ib_qp_open_attr {
1020 void (*event_handler)(struct ib_event *, void *);
1023 enum ib_qp_type qp_type;
1026 enum ib_rnr_timeout {
1027 IB_RNR_TIMER_655_36 = 0,
1028 IB_RNR_TIMER_000_01 = 1,
1029 IB_RNR_TIMER_000_02 = 2,
1030 IB_RNR_TIMER_000_03 = 3,
1031 IB_RNR_TIMER_000_04 = 4,
1032 IB_RNR_TIMER_000_06 = 5,
1033 IB_RNR_TIMER_000_08 = 6,
1034 IB_RNR_TIMER_000_12 = 7,
1035 IB_RNR_TIMER_000_16 = 8,
1036 IB_RNR_TIMER_000_24 = 9,
1037 IB_RNR_TIMER_000_32 = 10,
1038 IB_RNR_TIMER_000_48 = 11,
1039 IB_RNR_TIMER_000_64 = 12,
1040 IB_RNR_TIMER_000_96 = 13,
1041 IB_RNR_TIMER_001_28 = 14,
1042 IB_RNR_TIMER_001_92 = 15,
1043 IB_RNR_TIMER_002_56 = 16,
1044 IB_RNR_TIMER_003_84 = 17,
1045 IB_RNR_TIMER_005_12 = 18,
1046 IB_RNR_TIMER_007_68 = 19,
1047 IB_RNR_TIMER_010_24 = 20,
1048 IB_RNR_TIMER_015_36 = 21,
1049 IB_RNR_TIMER_020_48 = 22,
1050 IB_RNR_TIMER_030_72 = 23,
1051 IB_RNR_TIMER_040_96 = 24,
1052 IB_RNR_TIMER_061_44 = 25,
1053 IB_RNR_TIMER_081_92 = 26,
1054 IB_RNR_TIMER_122_88 = 27,
1055 IB_RNR_TIMER_163_84 = 28,
1056 IB_RNR_TIMER_245_76 = 29,
1057 IB_RNR_TIMER_327_68 = 30,
1058 IB_RNR_TIMER_491_52 = 31
1061 enum ib_qp_attr_mask {
1063 IB_QP_CUR_STATE = (1<<1),
1064 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1065 IB_QP_ACCESS_FLAGS = (1<<3),
1066 IB_QP_PKEY_INDEX = (1<<4),
1067 IB_QP_PORT = (1<<5),
1068 IB_QP_QKEY = (1<<6),
1070 IB_QP_PATH_MTU = (1<<8),
1071 IB_QP_TIMEOUT = (1<<9),
1072 IB_QP_RETRY_CNT = (1<<10),
1073 IB_QP_RNR_RETRY = (1<<11),
1074 IB_QP_RQ_PSN = (1<<12),
1075 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1076 IB_QP_ALT_PATH = (1<<14),
1077 IB_QP_MIN_RNR_TIMER = (1<<15),
1078 IB_QP_SQ_PSN = (1<<16),
1079 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1080 IB_QP_PATH_MIG_STATE = (1<<18),
1081 IB_QP_CAP = (1<<19),
1082 IB_QP_DEST_QPN = (1<<20),
1083 IB_QP_RESERVED1 = (1<<21),
1084 IB_QP_RESERVED2 = (1<<22),
1085 IB_QP_RESERVED3 = (1<<23),
1086 IB_QP_RESERVED4 = (1<<24),
1111 enum ib_qp_state qp_state;
1112 enum ib_qp_state cur_qp_state;
1113 enum ib_mtu path_mtu;
1114 enum ib_mig_state path_mig_state;
1119 int qp_access_flags;
1120 struct ib_qp_cap cap;
1121 struct ib_ah_attr ah_attr;
1122 struct ib_ah_attr alt_ah_attr;
1125 u8 en_sqd_async_notify;
1128 u8 max_dest_rd_atomic;
1140 IB_WR_RDMA_WRITE_WITH_IMM,
1142 IB_WR_SEND_WITH_IMM,
1144 IB_WR_ATOMIC_CMP_AND_SWP,
1145 IB_WR_ATOMIC_FETCH_AND_ADD,
1147 IB_WR_SEND_WITH_INV,
1148 IB_WR_RDMA_READ_WITH_INV,
1151 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1152 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1154 /* reserve values for low level drivers' internal use.
1155 * These values will not be used at all in the ib core layer.
1157 IB_WR_RESERVED1 = 0xf0,
1169 enum ib_send_flags {
1171 IB_SEND_SIGNALED = (1<<1),
1172 IB_SEND_SOLICITED = (1<<2),
1173 IB_SEND_INLINE = (1<<3),
1174 IB_SEND_IP_CSUM = (1<<4),
1176 /* reserve bits 26-31 for low level drivers' internal use */
1177 IB_SEND_RESERVED_START = (1 << 26),
1178 IB_SEND_RESERVED_END = (1 << 31),
1188 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1192 struct ib_send_wr *next;
1195 struct ib_cqe *wr_cqe;
1197 struct ib_sge *sg_list;
1199 enum ib_wr_opcode opcode;
1203 u32 invalidate_rkey;
1208 struct ib_send_wr wr;
1213 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1215 return container_of(wr, struct ib_rdma_wr, wr);
1218 struct ib_atomic_wr {
1219 struct ib_send_wr wr;
1223 u64 compare_add_mask;
1228 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1230 return container_of(wr, struct ib_atomic_wr, wr);
1234 struct ib_send_wr wr;
1241 u16 pkey_index; /* valid for GSI only */
1242 u8 port_num; /* valid for DR SMPs on switch only */
1245 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1247 return container_of(wr, struct ib_ud_wr, wr);
1251 struct ib_send_wr wr;
1257 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1259 return container_of(wr, struct ib_reg_wr, wr);
1262 struct ib_sig_handover_wr {
1263 struct ib_send_wr wr;
1264 struct ib_sig_attrs *sig_attrs;
1265 struct ib_mr *sig_mr;
1267 struct ib_sge *prot;
1270 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1272 return container_of(wr, struct ib_sig_handover_wr, wr);
1276 struct ib_recv_wr *next;
1279 struct ib_cqe *wr_cqe;
1281 struct ib_sge *sg_list;
1285 enum ib_access_flags {
1286 IB_ACCESS_LOCAL_WRITE = 1,
1287 IB_ACCESS_REMOTE_WRITE = (1<<1),
1288 IB_ACCESS_REMOTE_READ = (1<<2),
1289 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1290 IB_ACCESS_MW_BIND = (1<<4),
1291 IB_ZERO_BASED = (1<<5),
1292 IB_ACCESS_ON_DEMAND = (1<<6),
1296 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1297 * are hidden here instead of a uapi header!
1299 enum ib_mr_rereg_flags {
1300 IB_MR_REREG_TRANS = 1,
1301 IB_MR_REREG_PD = (1<<1),
1302 IB_MR_REREG_ACCESS = (1<<2),
1303 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1306 struct ib_fmr_attr {
1314 struct ib_ucontext {
1315 struct ib_device *device;
1316 struct list_head pd_list;
1317 struct list_head mr_list;
1318 struct list_head mw_list;
1319 struct list_head cq_list;
1320 struct list_head qp_list;
1321 struct list_head srq_list;
1322 struct list_head ah_list;
1323 struct list_head xrcd_list;
1324 struct list_head rule_list;
1328 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1329 struct rb_root umem_tree;
1331 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1332 * mmu notifiers registration.
1334 struct rw_semaphore umem_rwsem;
1335 void (*invalidate_range)(struct ib_umem *umem,
1336 unsigned long start, unsigned long end);
1338 struct mmu_notifier mn;
1339 atomic_t notifier_count;
1340 /* A list of umems that don't have private mmu notifier counters yet. */
1341 struct list_head no_private_counters;
1347 u64 user_handle; /* handle given to us by userspace */
1348 struct ib_ucontext *context; /* associated user context */
1349 void *object; /* containing object */
1350 struct list_head list; /* link to context's list */
1351 int id; /* index into kernel idr */
1353 struct rw_semaphore mutex; /* protects .live */
1354 struct rcu_head rcu; /* kfree_rcu() overhead */
1359 const void __user *inbuf;
1360 void __user *outbuf;
1367 struct ib_device *device;
1368 struct ib_uobject *uobject;
1369 atomic_t usecnt; /* count all resources */
1370 struct ib_mr *local_mr;
1374 struct ib_device *device;
1375 atomic_t usecnt; /* count all exposed resources */
1376 struct inode *inode;
1378 struct mutex tgt_qp_mutex;
1379 struct list_head tgt_qp_list;
1383 struct ib_device *device;
1385 struct ib_uobject *uobject;
1388 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1390 enum ib_poll_context {
1391 IB_POLL_DIRECT, /* caller context, no hw completions */
1392 IB_POLL_SOFTIRQ, /* poll from softirq context */
1393 IB_POLL_WORKQUEUE, /* poll from workqueue */
1397 struct ib_device *device;
1398 struct ib_uobject *uobject;
1399 ib_comp_handler comp_handler;
1400 void (*event_handler)(struct ib_event *, void *);
1403 atomic_t usecnt; /* count number of work queues */
1404 enum ib_poll_context poll_ctx;
1407 struct irq_poll iop;
1408 struct work_struct work;
1413 struct ib_device *device;
1415 struct ib_uobject *uobject;
1416 void (*event_handler)(struct ib_event *, void *);
1418 enum ib_srq_type srq_type;
1423 struct ib_xrcd *xrcd;
1431 struct ib_device *device;
1433 struct ib_cq *send_cq;
1434 struct ib_cq *recv_cq;
1437 struct list_head rdma_mrs;
1438 struct list_head sig_mrs;
1440 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1441 struct list_head xrcd_list;
1443 /* count times opened, mcast attaches, flow attaches */
1445 struct list_head open_list;
1446 struct ib_qp *real_qp;
1447 struct ib_uobject *uobject;
1448 void (*event_handler)(struct ib_event *, void *);
1451 enum ib_qp_type qp_type;
1455 struct ib_device *device;
1461 unsigned int page_size;
1464 struct ib_uobject *uobject; /* user */
1465 struct list_head qp_entry; /* FR */
1470 struct ib_device *device;
1472 struct ib_uobject *uobject;
1474 enum ib_mw_type type;
1478 struct ib_device *device;
1480 struct list_head list;
1485 /* Supported steering options */
1486 enum ib_flow_attr_type {
1487 /* steering according to rule specifications */
1488 IB_FLOW_ATTR_NORMAL = 0x0,
1489 /* default unicast and multicast rule -
1490 * receive all Eth traffic which isn't steered to any QP
1492 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1493 /* default multicast rule -
1494 * receive all Eth multicast traffic which isn't steered to any QP
1496 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1497 /* sniffer rule - receive all port traffic */
1498 IB_FLOW_ATTR_SNIFFER = 0x3
1501 /* Supported steering header types */
1502 enum ib_flow_spec_type {
1504 IB_FLOW_SPEC_ETH = 0x20,
1505 IB_FLOW_SPEC_IB = 0x22,
1507 IB_FLOW_SPEC_IPV4 = 0x30,
1509 IB_FLOW_SPEC_TCP = 0x40,
1510 IB_FLOW_SPEC_UDP = 0x41
1512 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1513 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1515 /* Flow steering rule priority is set according to it's domain.
1516 * Lower domain value means higher priority.
1518 enum ib_flow_domain {
1519 IB_FLOW_DOMAIN_USER,
1520 IB_FLOW_DOMAIN_ETHTOOL,
1523 IB_FLOW_DOMAIN_NUM /* Must be last */
1526 enum ib_flow_flags {
1527 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1528 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1531 struct ib_flow_eth_filter {
1538 struct ib_flow_spec_eth {
1539 enum ib_flow_spec_type type;
1541 struct ib_flow_eth_filter val;
1542 struct ib_flow_eth_filter mask;
1545 struct ib_flow_ib_filter {
1550 struct ib_flow_spec_ib {
1551 enum ib_flow_spec_type type;
1553 struct ib_flow_ib_filter val;
1554 struct ib_flow_ib_filter mask;
1557 struct ib_flow_ipv4_filter {
1562 struct ib_flow_spec_ipv4 {
1563 enum ib_flow_spec_type type;
1565 struct ib_flow_ipv4_filter val;
1566 struct ib_flow_ipv4_filter mask;
1569 struct ib_flow_tcp_udp_filter {
1574 struct ib_flow_spec_tcp_udp {
1575 enum ib_flow_spec_type type;
1577 struct ib_flow_tcp_udp_filter val;
1578 struct ib_flow_tcp_udp_filter mask;
1581 union ib_flow_spec {
1583 enum ib_flow_spec_type type;
1586 struct ib_flow_spec_eth eth;
1587 struct ib_flow_spec_ib ib;
1588 struct ib_flow_spec_ipv4 ipv4;
1589 struct ib_flow_spec_tcp_udp tcp_udp;
1592 struct ib_flow_attr {
1593 enum ib_flow_attr_type type;
1599 /* Following are the optional layers according to user request
1600 * struct ib_flow_spec_xxx
1601 * struct ib_flow_spec_yyy
1607 struct ib_uobject *uobject;
1613 enum ib_process_mad_flags {
1614 IB_MAD_IGNORE_MKEY = 1,
1615 IB_MAD_IGNORE_BKEY = 2,
1616 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1619 enum ib_mad_result {
1620 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1621 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1622 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1623 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1626 #define IB_DEVICE_NAME_MAX 64
1630 struct ib_event_handler event_handler;
1631 struct ib_pkey_cache **pkey_cache;
1632 struct ib_gid_table **gid_cache;
1636 struct ib_dma_mapping_ops {
1637 int (*mapping_error)(struct ib_device *dev,
1639 u64 (*map_single)(struct ib_device *dev,
1640 void *ptr, size_t size,
1641 enum dma_data_direction direction);
1642 void (*unmap_single)(struct ib_device *dev,
1643 u64 addr, size_t size,
1644 enum dma_data_direction direction);
1645 u64 (*map_page)(struct ib_device *dev,
1646 struct page *page, unsigned long offset,
1648 enum dma_data_direction direction);
1649 void (*unmap_page)(struct ib_device *dev,
1650 u64 addr, size_t size,
1651 enum dma_data_direction direction);
1652 int (*map_sg)(struct ib_device *dev,
1653 struct scatterlist *sg, int nents,
1654 enum dma_data_direction direction);
1655 void (*unmap_sg)(struct ib_device *dev,
1656 struct scatterlist *sg, int nents,
1657 enum dma_data_direction direction);
1658 void (*sync_single_for_cpu)(struct ib_device *dev,
1661 enum dma_data_direction dir);
1662 void (*sync_single_for_device)(struct ib_device *dev,
1665 enum dma_data_direction dir);
1666 void *(*alloc_coherent)(struct ib_device *dev,
1670 void (*free_coherent)(struct ib_device *dev,
1671 size_t size, void *cpu_addr,
1677 struct ib_port_immutable {
1685 struct device *dma_device;
1687 char name[IB_DEVICE_NAME_MAX];
1689 struct list_head event_handler_list;
1690 spinlock_t event_handler_lock;
1692 spinlock_t client_data_lock;
1693 struct list_head core_list;
1694 /* Access to the client_data_list is protected by the client_data_lock
1695 * spinlock and the lists_rwsem read-write semaphore */
1696 struct list_head client_data_list;
1698 struct ib_cache cache;
1700 * port_immutable is indexed by port number
1702 struct ib_port_immutable *port_immutable;
1704 int num_comp_vectors;
1706 struct iw_cm_verbs *iwcm;
1708 int (*get_protocol_stats)(struct ib_device *device,
1709 union rdma_protocol_stats *stats);
1710 int (*query_device)(struct ib_device *device,
1711 struct ib_device_attr *device_attr,
1712 struct ib_udata *udata);
1713 int (*query_port)(struct ib_device *device,
1715 struct ib_port_attr *port_attr);
1716 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1718 /* When calling get_netdev, the HW vendor's driver should return the
1719 * net device of device @device at port @port_num or NULL if such
1720 * a net device doesn't exist. The vendor driver should call dev_hold
1721 * on this net device. The HW vendor's device driver must guarantee
1722 * that this function returns NULL before the net device reaches
1723 * NETDEV_UNREGISTER_FINAL state.
1725 struct net_device *(*get_netdev)(struct ib_device *device,
1727 int (*query_gid)(struct ib_device *device,
1728 u8 port_num, int index,
1730 /* When calling add_gid, the HW vendor's driver should
1731 * add the gid of device @device at gid index @index of
1732 * port @port_num to be @gid. Meta-info of that gid (for example,
1733 * the network device related to this gid is available
1734 * at @attr. @context allows the HW vendor driver to store extra
1735 * information together with a GID entry. The HW vendor may allocate
1736 * memory to contain this information and store it in @context when a
1737 * new GID entry is written to. Params are consistent until the next
1738 * call of add_gid or delete_gid. The function should return 0 on
1739 * success or error otherwise. The function could be called
1740 * concurrently for different ports. This function is only called
1741 * when roce_gid_table is used.
1743 int (*add_gid)(struct ib_device *device,
1746 const union ib_gid *gid,
1747 const struct ib_gid_attr *attr,
1749 /* When calling del_gid, the HW vendor's driver should delete the
1750 * gid of device @device at gid index @index of port @port_num.
1751 * Upon the deletion of a GID entry, the HW vendor must free any
1752 * allocated memory. The caller will clear @context afterwards.
1753 * This function is only called when roce_gid_table is used.
1755 int (*del_gid)(struct ib_device *device,
1759 int (*query_pkey)(struct ib_device *device,
1760 u8 port_num, u16 index, u16 *pkey);
1761 int (*modify_device)(struct ib_device *device,
1762 int device_modify_mask,
1763 struct ib_device_modify *device_modify);
1764 int (*modify_port)(struct ib_device *device,
1765 u8 port_num, int port_modify_mask,
1766 struct ib_port_modify *port_modify);
1767 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1768 struct ib_udata *udata);
1769 int (*dealloc_ucontext)(struct ib_ucontext *context);
1770 int (*mmap)(struct ib_ucontext *context,
1771 struct vm_area_struct *vma);
1772 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1773 struct ib_ucontext *context,
1774 struct ib_udata *udata);
1775 int (*dealloc_pd)(struct ib_pd *pd);
1776 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1777 struct ib_ah_attr *ah_attr);
1778 int (*modify_ah)(struct ib_ah *ah,
1779 struct ib_ah_attr *ah_attr);
1780 int (*query_ah)(struct ib_ah *ah,
1781 struct ib_ah_attr *ah_attr);
1782 int (*destroy_ah)(struct ib_ah *ah);
1783 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1784 struct ib_srq_init_attr *srq_init_attr,
1785 struct ib_udata *udata);
1786 int (*modify_srq)(struct ib_srq *srq,
1787 struct ib_srq_attr *srq_attr,
1788 enum ib_srq_attr_mask srq_attr_mask,
1789 struct ib_udata *udata);
1790 int (*query_srq)(struct ib_srq *srq,
1791 struct ib_srq_attr *srq_attr);
1792 int (*destroy_srq)(struct ib_srq *srq);
1793 int (*post_srq_recv)(struct ib_srq *srq,
1794 struct ib_recv_wr *recv_wr,
1795 struct ib_recv_wr **bad_recv_wr);
1796 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1797 struct ib_qp_init_attr *qp_init_attr,
1798 struct ib_udata *udata);
1799 int (*modify_qp)(struct ib_qp *qp,
1800 struct ib_qp_attr *qp_attr,
1802 struct ib_udata *udata);
1803 int (*query_qp)(struct ib_qp *qp,
1804 struct ib_qp_attr *qp_attr,
1806 struct ib_qp_init_attr *qp_init_attr);
1807 int (*destroy_qp)(struct ib_qp *qp);
1808 int (*post_send)(struct ib_qp *qp,
1809 struct ib_send_wr *send_wr,
1810 struct ib_send_wr **bad_send_wr);
1811 int (*post_recv)(struct ib_qp *qp,
1812 struct ib_recv_wr *recv_wr,
1813 struct ib_recv_wr **bad_recv_wr);
1814 struct ib_cq * (*create_cq)(struct ib_device *device,
1815 const struct ib_cq_init_attr *attr,
1816 struct ib_ucontext *context,
1817 struct ib_udata *udata);
1818 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1820 int (*destroy_cq)(struct ib_cq *cq);
1821 int (*resize_cq)(struct ib_cq *cq, int cqe,
1822 struct ib_udata *udata);
1823 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1825 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1826 int (*req_notify_cq)(struct ib_cq *cq,
1827 enum ib_cq_notify_flags flags);
1828 int (*req_ncomp_notif)(struct ib_cq *cq,
1830 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1831 int mr_access_flags);
1832 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1833 u64 start, u64 length,
1835 int mr_access_flags,
1836 struct ib_udata *udata);
1837 int (*rereg_user_mr)(struct ib_mr *mr,
1839 u64 start, u64 length,
1841 int mr_access_flags,
1843 struct ib_udata *udata);
1844 int (*dereg_mr)(struct ib_mr *mr);
1845 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1846 enum ib_mr_type mr_type,
1848 int (*map_mr_sg)(struct ib_mr *mr,
1849 struct scatterlist *sg,
1851 unsigned sg_offset);
1852 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1853 enum ib_mw_type type,
1854 struct ib_udata *udata);
1855 int (*dealloc_mw)(struct ib_mw *mw);
1856 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1857 int mr_access_flags,
1858 struct ib_fmr_attr *fmr_attr);
1859 int (*map_phys_fmr)(struct ib_fmr *fmr,
1860 u64 *page_list, int list_len,
1862 int (*unmap_fmr)(struct list_head *fmr_list);
1863 int (*dealloc_fmr)(struct ib_fmr *fmr);
1864 int (*attach_mcast)(struct ib_qp *qp,
1867 int (*detach_mcast)(struct ib_qp *qp,
1870 int (*process_mad)(struct ib_device *device,
1871 int process_mad_flags,
1873 const struct ib_wc *in_wc,
1874 const struct ib_grh *in_grh,
1875 const struct ib_mad_hdr *in_mad,
1877 struct ib_mad_hdr *out_mad,
1878 size_t *out_mad_size,
1879 u16 *out_mad_pkey_index);
1880 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1881 struct ib_ucontext *ucontext,
1882 struct ib_udata *udata);
1883 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1884 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1888 int (*destroy_flow)(struct ib_flow *flow_id);
1889 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1890 struct ib_mr_status *mr_status);
1891 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1892 void (*drain_rq)(struct ib_qp *qp);
1893 void (*drain_sq)(struct ib_qp *qp);
1894 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1896 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1897 struct ifla_vf_info *ivf);
1898 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1899 struct ifla_vf_stats *stats);
1900 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1903 struct ib_dma_mapping_ops *dma_ops;
1905 struct module *owner;
1907 struct kobject *ports_parent;
1908 struct list_head port_list;
1911 IB_DEV_UNINITIALIZED,
1917 u64 uverbs_cmd_mask;
1918 u64 uverbs_ex_cmd_mask;
1926 struct ib_device_attr attrs;
1929 * The following mandatory functions are used only at device
1930 * registration. Keep functions such as these at the end of this
1931 * structure to avoid cache line misses when accessing struct ib_device
1934 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1939 void (*add) (struct ib_device *);
1940 void (*remove)(struct ib_device *, void *client_data);
1942 /* Returns the net_dev belonging to this ib_client and matching the
1944 * @dev: An RDMA device that the net_dev use for communication.
1945 * @port: A physical port number on the RDMA device.
1946 * @pkey: P_Key that the net_dev uses if applicable.
1947 * @gid: A GID that the net_dev uses to communicate.
1948 * @addr: An IP address the net_dev is configured with.
1949 * @client_data: The device's client data set by ib_set_client_data().
1951 * An ib_client that implements a net_dev on top of RDMA devices
1952 * (such as IP over IB) should implement this callback, allowing the
1953 * rdma_cm module to find the right net_dev for a given request.
1955 * The caller is responsible for calling dev_put on the returned
1957 struct net_device *(*get_net_dev_by_params)(
1958 struct ib_device *dev,
1961 const union ib_gid *gid,
1962 const struct sockaddr *addr,
1964 struct list_head list;
1967 struct ib_device *ib_alloc_device(size_t size);
1968 void ib_dealloc_device(struct ib_device *device);
1970 int ib_register_device(struct ib_device *device,
1971 int (*port_callback)(struct ib_device *,
1972 u8, struct kobject *));
1973 void ib_unregister_device(struct ib_device *device);
1975 int ib_register_client (struct ib_client *client);
1976 void ib_unregister_client(struct ib_client *client);
1978 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1979 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1982 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1984 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1987 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1989 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1992 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
1996 const void __user *p = udata->inbuf + offset;
2000 if (len > USHRT_MAX)
2003 buf = kmalloc(len, GFP_KERNEL);
2007 if (copy_from_user(buf, p, len))
2010 ret = !memchr_inv(buf, 0, len);
2018 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2019 * contains all required attributes and no attributes not allowed for
2020 * the given QP state transition.
2021 * @cur_state: Current QP state
2022 * @next_state: Next QP state
2024 * @mask: Mask of supplied QP attributes
2025 * @ll : link layer of port
2027 * This function is a helper function that a low-level driver's
2028 * modify_qp method can use to validate the consumer's input. It
2029 * checks that cur_state and next_state are valid QP states, that a
2030 * transition from cur_state to next_state is allowed by the IB spec,
2031 * and that the attribute mask supplied is allowed for the transition.
2033 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2034 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2035 enum rdma_link_layer ll);
2037 int ib_register_event_handler (struct ib_event_handler *event_handler);
2038 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2039 void ib_dispatch_event(struct ib_event *event);
2041 int ib_query_port(struct ib_device *device,
2042 u8 port_num, struct ib_port_attr *port_attr);
2044 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2048 * rdma_cap_ib_switch - Check if the device is IB switch
2049 * @device: Device to check
2051 * Device driver is responsible for setting is_switch bit on
2052 * in ib_device structure at init time.
2054 * Return: true if the device is IB switch.
2056 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2058 return device->is_switch;
2062 * rdma_start_port - Return the first valid port number for the device
2065 * @device: Device to be checked
2067 * Return start port number
2069 static inline u8 rdma_start_port(const struct ib_device *device)
2071 return rdma_cap_ib_switch(device) ? 0 : 1;
2075 * rdma_end_port - Return the last valid port number for the device
2078 * @device: Device to be checked
2080 * Return last port number
2082 static inline u8 rdma_end_port(const struct ib_device *device)
2084 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2087 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2089 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2092 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2094 return device->port_immutable[port_num].core_cap_flags &
2095 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2098 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2100 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2103 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2105 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2108 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2110 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2113 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2115 return rdma_protocol_ib(device, port_num) ||
2116 rdma_protocol_roce(device, port_num);
2120 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2121 * Management Datagrams.
2122 * @device: Device to check
2123 * @port_num: Port number to check
2125 * Management Datagrams (MAD) are a required part of the InfiniBand
2126 * specification and are supported on all InfiniBand devices. A slightly
2127 * extended version are also supported on OPA interfaces.
2129 * Return: true if the port supports sending/receiving of MAD packets.
2131 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2133 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2137 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2138 * Management Datagrams.
2139 * @device: Device to check
2140 * @port_num: Port number to check
2142 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2143 * datagrams with their own versions. These OPA MADs share many but not all of
2144 * the characteristics of InfiniBand MADs.
2146 * OPA MADs differ in the following ways:
2148 * 1) MADs are variable size up to 2K
2149 * IBTA defined MADs remain fixed at 256 bytes
2150 * 2) OPA SMPs must carry valid PKeys
2151 * 3) OPA SMP packets are a different format
2153 * Return: true if the port supports OPA MAD packet formats.
2155 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2157 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2158 == RDMA_CORE_CAP_OPA_MAD;
2162 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2163 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2164 * @device: Device to check
2165 * @port_num: Port number to check
2167 * Each InfiniBand node is required to provide a Subnet Management Agent
2168 * that the subnet manager can access. Prior to the fabric being fully
2169 * configured by the subnet manager, the SMA is accessed via a well known
2170 * interface called the Subnet Management Interface (SMI). This interface
2171 * uses directed route packets to communicate with the SM to get around the
2172 * chicken and egg problem of the SM needing to know what's on the fabric
2173 * in order to configure the fabric, and needing to configure the fabric in
2174 * order to send packets to the devices on the fabric. These directed
2175 * route packets do not need the fabric fully configured in order to reach
2176 * their destination. The SMI is the only method allowed to send
2177 * directed route packets on an InfiniBand fabric.
2179 * Return: true if the port provides an SMI.
2181 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2183 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2187 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2188 * Communication Manager.
2189 * @device: Device to check
2190 * @port_num: Port number to check
2192 * The InfiniBand Communication Manager is one of many pre-defined General
2193 * Service Agents (GSA) that are accessed via the General Service
2194 * Interface (GSI). It's role is to facilitate establishment of connections
2195 * between nodes as well as other management related tasks for established
2198 * Return: true if the port supports an IB CM (this does not guarantee that
2199 * a CM is actually running however).
2201 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2203 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2207 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2208 * Communication Manager.
2209 * @device: Device to check
2210 * @port_num: Port number to check
2212 * Similar to above, but specific to iWARP connections which have a different
2213 * managment protocol than InfiniBand.
2215 * Return: true if the port supports an iWARP CM (this does not guarantee that
2216 * a CM is actually running however).
2218 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2220 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2224 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2225 * Subnet Administration.
2226 * @device: Device to check
2227 * @port_num: Port number to check
2229 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2230 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2231 * fabrics, devices should resolve routes to other hosts by contacting the
2232 * SA to query the proper route.
2234 * Return: true if the port should act as a client to the fabric Subnet
2235 * Administration interface. This does not imply that the SA service is
2238 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2240 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2244 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2246 * @device: Device to check
2247 * @port_num: Port number to check
2249 * InfiniBand multicast registration is more complex than normal IPv4 or
2250 * IPv6 multicast registration. Each Host Channel Adapter must register
2251 * with the Subnet Manager when it wishes to join a multicast group. It
2252 * should do so only once regardless of how many queue pairs it subscribes
2253 * to this group. And it should leave the group only after all queue pairs
2254 * attached to the group have been detached.
2256 * Return: true if the port must undertake the additional adminstrative
2257 * overhead of registering/unregistering with the SM and tracking of the
2258 * total number of queue pairs attached to the multicast group.
2260 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2262 return rdma_cap_ib_sa(device, port_num);
2266 * rdma_cap_af_ib - Check if the port of device has the capability
2267 * Native Infiniband Address.
2268 * @device: Device to check
2269 * @port_num: Port number to check
2271 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2272 * GID. RoCE uses a different mechanism, but still generates a GID via
2273 * a prescribed mechanism and port specific data.
2275 * Return: true if the port uses a GID address to identify devices on the
2278 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2280 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2284 * rdma_cap_eth_ah - Check if the port of device has the capability
2285 * Ethernet Address Handle.
2286 * @device: Device to check
2287 * @port_num: Port number to check
2289 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2290 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2291 * port. Normally, packet headers are generated by the sending host
2292 * adapter, but when sending connectionless datagrams, we must manually
2293 * inject the proper headers for the fabric we are communicating over.
2295 * Return: true if we are running as a RoCE port and must force the
2296 * addition of a Global Route Header built from our Ethernet Address
2297 * Handle into our header list for connectionless packets.
2299 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2301 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2305 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2308 * @port_num: Port number
2310 * This MAD size includes the MAD headers and MAD payload. No other headers
2313 * Return the max MAD size required by the Port. Will return 0 if the port
2314 * does not support MADs
2316 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2318 return device->port_immutable[port_num].max_mad_size;
2322 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2323 * @device: Device to check
2324 * @port_num: Port number to check
2326 * RoCE GID table mechanism manages the various GIDs for a device.
2328 * NOTE: if allocating the port's GID table has failed, this call will still
2329 * return true, but any RoCE GID table API will fail.
2331 * Return: true if the port uses RoCE GID table mechanism in order to manage
2334 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2337 return rdma_protocol_roce(device, port_num) &&
2338 device->add_gid && device->del_gid;
2342 * Check if the device supports READ W/ INVALIDATE.
2344 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2347 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2348 * has support for it yet.
2350 return rdma_protocol_iwarp(dev, port_num);
2353 int ib_query_gid(struct ib_device *device,
2354 u8 port_num, int index, union ib_gid *gid,
2355 struct ib_gid_attr *attr);
2357 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2359 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2360 struct ifla_vf_info *info);
2361 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2362 struct ifla_vf_stats *stats);
2363 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2366 int ib_query_pkey(struct ib_device *device,
2367 u8 port_num, u16 index, u16 *pkey);
2369 int ib_modify_device(struct ib_device *device,
2370 int device_modify_mask,
2371 struct ib_device_modify *device_modify);
2373 int ib_modify_port(struct ib_device *device,
2374 u8 port_num, int port_modify_mask,
2375 struct ib_port_modify *port_modify);
2377 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2378 enum ib_gid_type gid_type, struct net_device *ndev,
2379 u8 *port_num, u16 *index);
2381 int ib_find_pkey(struct ib_device *device,
2382 u8 port_num, u16 pkey, u16 *index);
2384 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2386 void ib_dealloc_pd(struct ib_pd *pd);
2389 * ib_create_ah - Creates an address handle for the given address vector.
2390 * @pd: The protection domain associated with the address handle.
2391 * @ah_attr: The attributes of the address vector.
2393 * The address handle is used to reference a local or global destination
2394 * in all UD QP post sends.
2396 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2399 * ib_init_ah_from_wc - Initializes address handle attributes from a
2401 * @device: Device on which the received message arrived.
2402 * @port_num: Port on which the received message arrived.
2403 * @wc: Work completion associated with the received message.
2404 * @grh: References the received global route header. This parameter is
2405 * ignored unless the work completion indicates that the GRH is valid.
2406 * @ah_attr: Returned attributes that can be used when creating an address
2407 * handle for replying to the message.
2409 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2410 const struct ib_wc *wc, const struct ib_grh *grh,
2411 struct ib_ah_attr *ah_attr);
2414 * ib_create_ah_from_wc - Creates an address handle associated with the
2415 * sender of the specified work completion.
2416 * @pd: The protection domain associated with the address handle.
2417 * @wc: Work completion information associated with a received message.
2418 * @grh: References the received global route header. This parameter is
2419 * ignored unless the work completion indicates that the GRH is valid.
2420 * @port_num: The outbound port number to associate with the address.
2422 * The address handle is used to reference a local or global destination
2423 * in all UD QP post sends.
2425 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2426 const struct ib_grh *grh, u8 port_num);
2429 * ib_modify_ah - Modifies the address vector associated with an address
2431 * @ah: The address handle to modify.
2432 * @ah_attr: The new address vector attributes to associate with the
2435 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2438 * ib_query_ah - Queries the address vector associated with an address
2440 * @ah: The address handle to query.
2441 * @ah_attr: The address vector attributes associated with the address
2444 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2447 * ib_destroy_ah - Destroys an address handle.
2448 * @ah: The address handle to destroy.
2450 int ib_destroy_ah(struct ib_ah *ah);
2453 * ib_create_srq - Creates a SRQ associated with the specified protection
2455 * @pd: The protection domain associated with the SRQ.
2456 * @srq_init_attr: A list of initial attributes required to create the
2457 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2458 * the actual capabilities of the created SRQ.
2460 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2461 * requested size of the SRQ, and set to the actual values allocated
2462 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2463 * will always be at least as large as the requested values.
2465 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2466 struct ib_srq_init_attr *srq_init_attr);
2469 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2470 * @srq: The SRQ to modify.
2471 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2472 * the current values of selected SRQ attributes are returned.
2473 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2474 * are being modified.
2476 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2477 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2478 * the number of receives queued drops below the limit.
2480 int ib_modify_srq(struct ib_srq *srq,
2481 struct ib_srq_attr *srq_attr,
2482 enum ib_srq_attr_mask srq_attr_mask);
2485 * ib_query_srq - Returns the attribute list and current values for the
2487 * @srq: The SRQ to query.
2488 * @srq_attr: The attributes of the specified SRQ.
2490 int ib_query_srq(struct ib_srq *srq,
2491 struct ib_srq_attr *srq_attr);
2494 * ib_destroy_srq - Destroys the specified SRQ.
2495 * @srq: The SRQ to destroy.
2497 int ib_destroy_srq(struct ib_srq *srq);
2500 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2501 * @srq: The SRQ to post the work request on.
2502 * @recv_wr: A list of work requests to post on the receive queue.
2503 * @bad_recv_wr: On an immediate failure, this parameter will reference
2504 * the work request that failed to be posted on the QP.
2506 static inline int ib_post_srq_recv(struct ib_srq *srq,
2507 struct ib_recv_wr *recv_wr,
2508 struct ib_recv_wr **bad_recv_wr)
2510 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2514 * ib_create_qp - Creates a QP associated with the specified protection
2516 * @pd: The protection domain associated with the QP.
2517 * @qp_init_attr: A list of initial attributes required to create the
2518 * QP. If QP creation succeeds, then the attributes are updated to
2519 * the actual capabilities of the created QP.
2521 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2522 struct ib_qp_init_attr *qp_init_attr);
2525 * ib_modify_qp - Modifies the attributes for the specified QP and then
2526 * transitions the QP to the given state.
2527 * @qp: The QP to modify.
2528 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2529 * the current values of selected QP attributes are returned.
2530 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2531 * are being modified.
2533 int ib_modify_qp(struct ib_qp *qp,
2534 struct ib_qp_attr *qp_attr,
2538 * ib_query_qp - Returns the attribute list and current values for the
2540 * @qp: The QP to query.
2541 * @qp_attr: The attributes of the specified QP.
2542 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2543 * @qp_init_attr: Additional attributes of the selected QP.
2545 * The qp_attr_mask may be used to limit the query to gathering only the
2546 * selected attributes.
2548 int ib_query_qp(struct ib_qp *qp,
2549 struct ib_qp_attr *qp_attr,
2551 struct ib_qp_init_attr *qp_init_attr);
2554 * ib_destroy_qp - Destroys the specified QP.
2555 * @qp: The QP to destroy.
2557 int ib_destroy_qp(struct ib_qp *qp);
2560 * ib_open_qp - Obtain a reference to an existing sharable QP.
2561 * @xrcd - XRC domain
2562 * @qp_open_attr: Attributes identifying the QP to open.
2564 * Returns a reference to a sharable QP.
2566 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2567 struct ib_qp_open_attr *qp_open_attr);
2570 * ib_close_qp - Release an external reference to a QP.
2571 * @qp: The QP handle to release
2573 * The opened QP handle is released by the caller. The underlying
2574 * shared QP is not destroyed until all internal references are released.
2576 int ib_close_qp(struct ib_qp *qp);
2579 * ib_post_send - Posts a list of work requests to the send queue of
2581 * @qp: The QP to post the work request on.
2582 * @send_wr: A list of work requests to post on the send queue.
2583 * @bad_send_wr: On an immediate failure, this parameter will reference
2584 * the work request that failed to be posted on the QP.
2586 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2587 * error is returned, the QP state shall not be affected,
2588 * ib_post_send() will return an immediate error after queueing any
2589 * earlier work requests in the list.
2591 static inline int ib_post_send(struct ib_qp *qp,
2592 struct ib_send_wr *send_wr,
2593 struct ib_send_wr **bad_send_wr)
2595 return qp->device->post_send(qp, send_wr, bad_send_wr);
2599 * ib_post_recv - Posts a list of work requests to the receive queue of
2601 * @qp: The QP to post the work request on.
2602 * @recv_wr: A list of work requests to post on the receive queue.
2603 * @bad_recv_wr: On an immediate failure, this parameter will reference
2604 * the work request that failed to be posted on the QP.
2606 static inline int ib_post_recv(struct ib_qp *qp,
2607 struct ib_recv_wr *recv_wr,
2608 struct ib_recv_wr **bad_recv_wr)
2610 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2613 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2614 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2615 void ib_free_cq(struct ib_cq *cq);
2616 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2619 * ib_create_cq - Creates a CQ on the specified device.
2620 * @device: The device on which to create the CQ.
2621 * @comp_handler: A user-specified callback that is invoked when a
2622 * completion event occurs on the CQ.
2623 * @event_handler: A user-specified callback that is invoked when an
2624 * asynchronous event not associated with a completion occurs on the CQ.
2625 * @cq_context: Context associated with the CQ returned to the user via
2626 * the associated completion and event handlers.
2627 * @cq_attr: The attributes the CQ should be created upon.
2629 * Users can examine the cq structure to determine the actual CQ size.
2631 struct ib_cq *ib_create_cq(struct ib_device *device,
2632 ib_comp_handler comp_handler,
2633 void (*event_handler)(struct ib_event *, void *),
2635 const struct ib_cq_init_attr *cq_attr);
2638 * ib_resize_cq - Modifies the capacity of the CQ.
2639 * @cq: The CQ to resize.
2640 * @cqe: The minimum size of the CQ.
2642 * Users can examine the cq structure to determine the actual CQ size.
2644 int ib_resize_cq(struct ib_cq *cq, int cqe);
2647 * ib_modify_cq - Modifies moderation params of the CQ
2648 * @cq: The CQ to modify.
2649 * @cq_count: number of CQEs that will trigger an event
2650 * @cq_period: max period of time in usec before triggering an event
2653 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2656 * ib_destroy_cq - Destroys the specified CQ.
2657 * @cq: The CQ to destroy.
2659 int ib_destroy_cq(struct ib_cq *cq);
2662 * ib_poll_cq - poll a CQ for completion(s)
2663 * @cq:the CQ being polled
2664 * @num_entries:maximum number of completions to return
2665 * @wc:array of at least @num_entries &struct ib_wc where completions
2668 * Poll a CQ for (possibly multiple) completions. If the return value
2669 * is < 0, an error occurred. If the return value is >= 0, it is the
2670 * number of completions returned. If the return value is
2671 * non-negative and < num_entries, then the CQ was emptied.
2673 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2676 return cq->device->poll_cq(cq, num_entries, wc);
2680 * ib_peek_cq - Returns the number of unreaped completions currently
2681 * on the specified CQ.
2682 * @cq: The CQ to peek.
2683 * @wc_cnt: A minimum number of unreaped completions to check for.
2685 * If the number of unreaped completions is greater than or equal to wc_cnt,
2686 * this function returns wc_cnt, otherwise, it returns the actual number of
2687 * unreaped completions.
2689 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2692 * ib_req_notify_cq - Request completion notification on a CQ.
2693 * @cq: The CQ to generate an event for.
2695 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2696 * to request an event on the next solicited event or next work
2697 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2698 * may also be |ed in to request a hint about missed events, as
2702 * < 0 means an error occurred while requesting notification
2703 * == 0 means notification was requested successfully, and if
2704 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2705 * were missed and it is safe to wait for another event. In
2706 * this case is it guaranteed that any work completions added
2707 * to the CQ since the last CQ poll will trigger a completion
2708 * notification event.
2709 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2710 * in. It means that the consumer must poll the CQ again to
2711 * make sure it is empty to avoid missing an event because of a
2712 * race between requesting notification and an entry being
2713 * added to the CQ. This return value means it is possible
2714 * (but not guaranteed) that a work completion has been added
2715 * to the CQ since the last poll without triggering a
2716 * completion notification event.
2718 static inline int ib_req_notify_cq(struct ib_cq *cq,
2719 enum ib_cq_notify_flags flags)
2721 return cq->device->req_notify_cq(cq, flags);
2725 * ib_req_ncomp_notif - Request completion notification when there are
2726 * at least the specified number of unreaped completions on the CQ.
2727 * @cq: The CQ to generate an event for.
2728 * @wc_cnt: The number of unreaped completions that should be on the
2729 * CQ before an event is generated.
2731 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2733 return cq->device->req_ncomp_notif ?
2734 cq->device->req_ncomp_notif(cq, wc_cnt) :
2739 * ib_get_dma_mr - Returns a memory region for system memory that is
2741 * @pd: The protection domain associated with the memory region.
2742 * @mr_access_flags: Specifies the memory access rights.
2744 * Note that the ib_dma_*() functions defined below must be used
2745 * to create/destroy addresses used with the Lkey or Rkey returned
2746 * by ib_get_dma_mr().
2748 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2751 * ib_dma_mapping_error - check a DMA addr for error
2752 * @dev: The device for which the dma_addr was created
2753 * @dma_addr: The DMA address to check
2755 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2758 return dev->dma_ops->mapping_error(dev, dma_addr);
2759 return dma_mapping_error(dev->dma_device, dma_addr);
2763 * ib_dma_map_single - Map a kernel virtual address to DMA address
2764 * @dev: The device for which the dma_addr is to be created
2765 * @cpu_addr: The kernel virtual address
2766 * @size: The size of the region in bytes
2767 * @direction: The direction of the DMA
2769 static inline u64 ib_dma_map_single(struct ib_device *dev,
2770 void *cpu_addr, size_t size,
2771 enum dma_data_direction direction)
2774 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2775 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2779 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2780 * @dev: The device for which the DMA address was created
2781 * @addr: The DMA address
2782 * @size: The size of the region in bytes
2783 * @direction: The direction of the DMA
2785 static inline void ib_dma_unmap_single(struct ib_device *dev,
2786 u64 addr, size_t size,
2787 enum dma_data_direction direction)
2790 dev->dma_ops->unmap_single(dev, addr, size, direction);
2792 dma_unmap_single(dev->dma_device, addr, size, direction);
2795 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2796 void *cpu_addr, size_t size,
2797 enum dma_data_direction direction,
2798 struct dma_attrs *attrs)
2800 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2804 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2805 u64 addr, size_t size,
2806 enum dma_data_direction direction,
2807 struct dma_attrs *attrs)
2809 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2814 * ib_dma_map_page - Map a physical page to DMA address
2815 * @dev: The device for which the dma_addr is to be created
2816 * @page: The page to be mapped
2817 * @offset: The offset within the page
2818 * @size: The size of the region in bytes
2819 * @direction: The direction of the DMA
2821 static inline u64 ib_dma_map_page(struct ib_device *dev,
2823 unsigned long offset,
2825 enum dma_data_direction direction)
2828 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2829 return dma_map_page(dev->dma_device, page, offset, size, direction);
2833 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2834 * @dev: The device for which the DMA address was created
2835 * @addr: The DMA address
2836 * @size: The size of the region in bytes
2837 * @direction: The direction of the DMA
2839 static inline void ib_dma_unmap_page(struct ib_device *dev,
2840 u64 addr, size_t size,
2841 enum dma_data_direction direction)
2844 dev->dma_ops->unmap_page(dev, addr, size, direction);
2846 dma_unmap_page(dev->dma_device, addr, size, direction);
2850 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2851 * @dev: The device for which the DMA addresses are to be created
2852 * @sg: The array of scatter/gather entries
2853 * @nents: The number of scatter/gather entries
2854 * @direction: The direction of the DMA
2856 static inline int ib_dma_map_sg(struct ib_device *dev,
2857 struct scatterlist *sg, int nents,
2858 enum dma_data_direction direction)
2861 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2862 return dma_map_sg(dev->dma_device, sg, nents, direction);
2866 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2867 * @dev: The device for which the DMA addresses were created
2868 * @sg: The array of scatter/gather entries
2869 * @nents: The number of scatter/gather entries
2870 * @direction: The direction of the DMA
2872 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2873 struct scatterlist *sg, int nents,
2874 enum dma_data_direction direction)
2877 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2879 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2882 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2883 struct scatterlist *sg, int nents,
2884 enum dma_data_direction direction,
2885 struct dma_attrs *attrs)
2887 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2890 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2891 struct scatterlist *sg, int nents,
2892 enum dma_data_direction direction,
2893 struct dma_attrs *attrs)
2895 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2898 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2899 * @dev: The device for which the DMA addresses were created
2900 * @sg: The scatter/gather entry
2902 * Note: this function is obsolete. To do: change all occurrences of
2903 * ib_sg_dma_address() into sg_dma_address().
2905 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2906 struct scatterlist *sg)
2908 return sg_dma_address(sg);
2912 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2913 * @dev: The device for which the DMA addresses were created
2914 * @sg: The scatter/gather entry
2916 * Note: this function is obsolete. To do: change all occurrences of
2917 * ib_sg_dma_len() into sg_dma_len().
2919 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2920 struct scatterlist *sg)
2922 return sg_dma_len(sg);
2926 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2927 * @dev: The device for which the DMA address was created
2928 * @addr: The DMA address
2929 * @size: The size of the region in bytes
2930 * @dir: The direction of the DMA
2932 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2935 enum dma_data_direction dir)
2938 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2940 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2944 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2945 * @dev: The device for which the DMA address was created
2946 * @addr: The DMA address
2947 * @size: The size of the region in bytes
2948 * @dir: The direction of the DMA
2950 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2953 enum dma_data_direction dir)
2956 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2958 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2962 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2963 * @dev: The device for which the DMA address is requested
2964 * @size: The size of the region to allocate in bytes
2965 * @dma_handle: A pointer for returning the DMA address of the region
2966 * @flag: memory allocator flags
2968 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2974 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2979 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2980 *dma_handle = handle;
2986 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2987 * @dev: The device for which the DMA addresses were allocated
2988 * @size: The size of the region
2989 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2990 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2992 static inline void ib_dma_free_coherent(struct ib_device *dev,
2993 size_t size, void *cpu_addr,
2997 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2999 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3003 * ib_dereg_mr - Deregisters a memory region and removes it from the
3004 * HCA translation table.
3005 * @mr: The memory region to deregister.
3007 * This function can fail, if the memory region has memory windows bound to it.
3009 int ib_dereg_mr(struct ib_mr *mr);
3011 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3012 enum ib_mr_type mr_type,
3016 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3018 * @mr - struct ib_mr pointer to be updated.
3019 * @newkey - new key to be used.
3021 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3023 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3024 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3028 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3029 * for calculating a new rkey for type 2 memory windows.
3030 * @rkey - the rkey to increment.
3032 static inline u32 ib_inc_rkey(u32 rkey)
3034 const u32 mask = 0x000000ff;
3035 return ((rkey + 1) & mask) | (rkey & ~mask);
3039 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3040 * @pd: The protection domain associated with the unmapped region.
3041 * @mr_access_flags: Specifies the memory access rights.
3042 * @fmr_attr: Attributes of the unmapped region.
3044 * A fast memory region must be mapped before it can be used as part of
3047 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3048 int mr_access_flags,
3049 struct ib_fmr_attr *fmr_attr);
3052 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3053 * @fmr: The fast memory region to associate with the pages.
3054 * @page_list: An array of physical pages to map to the fast memory region.
3055 * @list_len: The number of pages in page_list.
3056 * @iova: The I/O virtual address to use with the mapped region.
3058 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3059 u64 *page_list, int list_len,
3062 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3066 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3067 * @fmr_list: A linked list of fast memory regions to unmap.
3069 int ib_unmap_fmr(struct list_head *fmr_list);
3072 * ib_dealloc_fmr - Deallocates a fast memory region.
3073 * @fmr: The fast memory region to deallocate.
3075 int ib_dealloc_fmr(struct ib_fmr *fmr);
3078 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3079 * @qp: QP to attach to the multicast group. The QP must be type
3081 * @gid: Multicast group GID.
3082 * @lid: Multicast group LID in host byte order.
3084 * In order to send and receive multicast packets, subnet
3085 * administration must have created the multicast group and configured
3086 * the fabric appropriately. The port associated with the specified
3087 * QP must also be a member of the multicast group.
3089 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3092 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3093 * @qp: QP to detach from the multicast group.
3094 * @gid: Multicast group GID.
3095 * @lid: Multicast group LID in host byte order.
3097 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3100 * ib_alloc_xrcd - Allocates an XRC domain.
3101 * @device: The device on which to allocate the XRC domain.
3103 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3106 * ib_dealloc_xrcd - Deallocates an XRC domain.
3107 * @xrcd: The XRC domain to deallocate.
3109 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3111 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3112 struct ib_flow_attr *flow_attr, int domain);
3113 int ib_destroy_flow(struct ib_flow *flow_id);
3115 static inline int ib_check_mr_access(int flags)
3118 * Local write permission is required if remote write or
3119 * remote atomic permission is also requested.
3121 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3122 !(flags & IB_ACCESS_LOCAL_WRITE))
3129 * ib_check_mr_status: lightweight check of MR status.
3130 * This routine may provide status checks on a selected
3131 * ib_mr. first use is for signature status check.
3133 * @mr: A memory region.
3134 * @check_mask: Bitmask of which checks to perform from
3135 * ib_mr_status_check enumeration.
3136 * @mr_status: The container of relevant status checks.
3137 * failed checks will be indicated in the status bitmask
3138 * and the relevant info shall be in the error item.
3140 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3141 struct ib_mr_status *mr_status);
3143 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3144 u16 pkey, const union ib_gid *gid,
3145 const struct sockaddr *addr);
3147 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3148 unsigned int sg_offset, unsigned int page_size);
3151 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3152 unsigned int sg_offset, unsigned int page_size)
3156 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3162 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3163 unsigned int sg_offset, int (*set_page)(struct ib_mr *, u64));
3165 void ib_drain_rq(struct ib_qp *qp);
3166 void ib_drain_sq(struct ib_qp *qp);
3167 void ib_drain_qp(struct ib_qp *qp);
3168 #endif /* IB_VERBS_H */