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),
223 IB_DEVICE_RAW_SCATTER_FCS = ((u64)1 << 34),
226 enum ib_signature_prot_cap {
227 IB_PROT_T10DIF_TYPE_1 = 1,
228 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
232 enum ib_signature_guard_cap {
233 IB_GUARD_T10DIF_CRC = 1,
234 IB_GUARD_T10DIF_CSUM = 1 << 1,
243 enum ib_odp_general_cap_bits {
244 IB_ODP_SUPPORT = 1 << 0,
247 enum ib_odp_transport_cap_bits {
248 IB_ODP_SUPPORT_SEND = 1 << 0,
249 IB_ODP_SUPPORT_RECV = 1 << 1,
250 IB_ODP_SUPPORT_WRITE = 1 << 2,
251 IB_ODP_SUPPORT_READ = 1 << 3,
252 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
256 uint64_t general_caps;
258 uint32_t rc_odp_caps;
259 uint32_t uc_odp_caps;
260 uint32_t ud_odp_caps;
261 } per_transport_caps;
264 enum ib_cq_creation_flags {
265 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
266 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
269 struct ib_cq_init_attr {
275 struct ib_device_attr {
277 __be64 sys_image_guid;
285 u64 device_cap_flags;
295 int max_qp_init_rd_atom;
296 int max_ee_init_rd_atom;
297 enum ib_atomic_cap atomic_cap;
298 enum ib_atomic_cap masked_atomic_cap;
305 int max_mcast_qp_attach;
306 int max_total_mcast_qp_attach;
313 unsigned int max_fast_reg_page_list_len;
315 u8 local_ca_ack_delay;
318 struct ib_odp_caps odp_caps;
319 uint64_t timestamp_mask;
320 uint64_t hca_core_clock; /* in KHZ */
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
334 case IB_MTU_256: return 256;
335 case IB_MTU_512: return 512;
336 case IB_MTU_1024: return 1024;
337 case IB_MTU_2048: return 2048;
338 case IB_MTU_4096: return 4096;
349 IB_PORT_ACTIVE_DEFER = 5
352 enum ib_port_cap_flags {
354 IB_PORT_NOTICE_SUP = 1 << 2,
355 IB_PORT_TRAP_SUP = 1 << 3,
356 IB_PORT_OPT_IPD_SUP = 1 << 4,
357 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
358 IB_PORT_SL_MAP_SUP = 1 << 6,
359 IB_PORT_MKEY_NVRAM = 1 << 7,
360 IB_PORT_PKEY_NVRAM = 1 << 8,
361 IB_PORT_LED_INFO_SUP = 1 << 9,
362 IB_PORT_SM_DISABLED = 1 << 10,
363 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
364 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
365 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
366 IB_PORT_CM_SUP = 1 << 16,
367 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
368 IB_PORT_REINIT_SUP = 1 << 18,
369 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
370 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
371 IB_PORT_DR_NOTICE_SUP = 1 << 21,
372 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
373 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
374 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
375 IB_PORT_CLIENT_REG_SUP = 1 << 25,
376 IB_PORT_IP_BASED_GIDS = 1 << 26,
386 static inline int ib_width_enum_to_int(enum ib_port_width width)
389 case IB_WIDTH_1X: return 1;
390 case IB_WIDTH_4X: return 4;
391 case IB_WIDTH_8X: return 8;
392 case IB_WIDTH_12X: return 12;
406 struct ib_protocol_stats {
410 struct iw_protocol_stats {
413 u64 ipInTooBigErrors;
416 u64 ipInUnknownProtos;
417 u64 ipInTruncatedPkts;
420 u64 ipOutForwDatagrams;
452 union rdma_protocol_stats {
453 struct ib_protocol_stats ib;
454 struct iw_protocol_stats iw;
457 /* Define bits for the various functionality this port needs to be supported by
460 /* Management 0x00000FFF */
461 #define RDMA_CORE_CAP_IB_MAD 0x00000001
462 #define RDMA_CORE_CAP_IB_SMI 0x00000002
463 #define RDMA_CORE_CAP_IB_CM 0x00000004
464 #define RDMA_CORE_CAP_IW_CM 0x00000008
465 #define RDMA_CORE_CAP_IB_SA 0x00000010
466 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
468 /* Address format 0x000FF000 */
469 #define RDMA_CORE_CAP_AF_IB 0x00001000
470 #define RDMA_CORE_CAP_ETH_AH 0x00002000
472 /* Protocol 0xFFF00000 */
473 #define RDMA_CORE_CAP_PROT_IB 0x00100000
474 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
475 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
476 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
478 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
479 | RDMA_CORE_CAP_IB_MAD \
480 | RDMA_CORE_CAP_IB_SMI \
481 | RDMA_CORE_CAP_IB_CM \
482 | RDMA_CORE_CAP_IB_SA \
483 | RDMA_CORE_CAP_AF_IB)
484 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
485 | RDMA_CORE_CAP_IB_MAD \
486 | RDMA_CORE_CAP_IB_CM \
487 | RDMA_CORE_CAP_AF_IB \
488 | RDMA_CORE_CAP_ETH_AH)
489 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
490 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
491 | RDMA_CORE_CAP_IB_MAD \
492 | RDMA_CORE_CAP_IB_CM \
493 | RDMA_CORE_CAP_AF_IB \
494 | RDMA_CORE_CAP_ETH_AH)
495 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
496 | RDMA_CORE_CAP_IW_CM)
497 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
498 | RDMA_CORE_CAP_OPA_MAD)
500 struct ib_port_attr {
502 enum ib_port_state state;
504 enum ib_mtu active_mtu;
524 enum ib_device_modify_flags {
525 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
526 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
529 struct ib_device_modify {
534 enum ib_port_modify_flags {
535 IB_PORT_SHUTDOWN = 1,
536 IB_PORT_INIT_TYPE = (1<<2),
537 IB_PORT_RESET_QKEY_CNTR = (1<<3)
540 struct ib_port_modify {
541 u32 set_port_cap_mask;
542 u32 clr_port_cap_mask;
550 IB_EVENT_QP_ACCESS_ERR,
554 IB_EVENT_PATH_MIG_ERR,
555 IB_EVENT_DEVICE_FATAL,
556 IB_EVENT_PORT_ACTIVE,
559 IB_EVENT_PKEY_CHANGE,
562 IB_EVENT_SRQ_LIMIT_REACHED,
563 IB_EVENT_QP_LAST_WQE_REACHED,
564 IB_EVENT_CLIENT_REREGISTER,
568 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
571 struct ib_device *device;
578 enum ib_event_type event;
581 struct ib_event_handler {
582 struct ib_device *device;
583 void (*handler)(struct ib_event_handler *, struct ib_event *);
584 struct list_head list;
587 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
589 (_ptr)->device = _device; \
590 (_ptr)->handler = _handler; \
591 INIT_LIST_HEAD(&(_ptr)->list); \
594 struct ib_global_route {
603 __be32 version_tclass_flow;
611 union rdma_network_hdr {
614 /* The IB spec states that if it's IPv4, the header
615 * is located in the last 20 bytes of the header.
618 struct iphdr roce4grh;
623 IB_MULTICAST_QPN = 0xffffff
626 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
627 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
634 IB_RATE_PORT_CURRENT = 0,
635 IB_RATE_2_5_GBPS = 2,
643 IB_RATE_120_GBPS = 10,
644 IB_RATE_14_GBPS = 11,
645 IB_RATE_56_GBPS = 12,
646 IB_RATE_112_GBPS = 13,
647 IB_RATE_168_GBPS = 14,
648 IB_RATE_25_GBPS = 15,
649 IB_RATE_100_GBPS = 16,
650 IB_RATE_200_GBPS = 17,
651 IB_RATE_300_GBPS = 18
655 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
656 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
657 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
658 * @rate: rate to convert.
660 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
663 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
664 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
665 * @rate: rate to convert.
667 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
671 * enum ib_mr_type - memory region type
672 * @IB_MR_TYPE_MEM_REG: memory region that is used for
673 * normal registration
674 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
675 * signature operations (data-integrity
677 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
678 * register any arbitrary sg lists (without
679 * the normal mr constraints - see
684 IB_MR_TYPE_SIGNATURE,
690 * IB_SIG_TYPE_NONE: Unprotected.
691 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
693 enum ib_signature_type {
699 * Signature T10-DIF block-guard types
700 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
701 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
703 enum ib_t10_dif_bg_type {
709 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
711 * @bg_type: T10-DIF block guard type (CRC|CSUM)
712 * @pi_interval: protection information interval.
713 * @bg: seed of guard computation.
714 * @app_tag: application tag of guard block
715 * @ref_tag: initial guard block reference tag.
716 * @ref_remap: Indicate wethear the reftag increments each block
717 * @app_escape: Indicate to skip block check if apptag=0xffff
718 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
719 * @apptag_check_mask: check bitmask of application tag.
721 struct ib_t10_dif_domain {
722 enum ib_t10_dif_bg_type bg_type;
730 u16 apptag_check_mask;
734 * struct ib_sig_domain - Parameters for signature domain
735 * @sig_type: specific signauture type
736 * @sig: union of all signature domain attributes that may
737 * be used to set domain layout.
739 struct ib_sig_domain {
740 enum ib_signature_type sig_type;
742 struct ib_t10_dif_domain dif;
747 * struct ib_sig_attrs - Parameters for signature handover operation
748 * @check_mask: bitmask for signature byte check (8 bytes)
749 * @mem: memory domain layout desciptor.
750 * @wire: wire domain layout desciptor.
752 struct ib_sig_attrs {
754 struct ib_sig_domain mem;
755 struct ib_sig_domain wire;
758 enum ib_sig_err_type {
765 * struct ib_sig_err - signature error descriptor
768 enum ib_sig_err_type err_type;
775 enum ib_mr_status_check {
776 IB_MR_CHECK_SIG_STATUS = 1,
780 * struct ib_mr_status - Memory region status container
782 * @fail_status: Bitmask of MR checks status. For each
783 * failed check a corresponding status bit is set.
784 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
787 struct ib_mr_status {
789 struct ib_sig_err sig_err;
793 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
795 * @mult: multiple to convert.
797 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
800 struct ib_global_route grh;
814 IB_WC_LOC_EEC_OP_ERR,
819 IB_WC_LOC_ACCESS_ERR,
820 IB_WC_REM_INV_REQ_ERR,
821 IB_WC_REM_ACCESS_ERR,
824 IB_WC_RNR_RETRY_EXC_ERR,
825 IB_WC_LOC_RDD_VIOL_ERR,
826 IB_WC_REM_INV_RD_REQ_ERR,
829 IB_WC_INV_EEC_STATE_ERR,
831 IB_WC_RESP_TIMEOUT_ERR,
835 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
846 IB_WC_MASKED_COMP_SWAP,
847 IB_WC_MASKED_FETCH_ADD,
849 * Set value of IB_WC_RECV so consumers can test if a completion is a
850 * receive by testing (opcode & IB_WC_RECV).
853 IB_WC_RECV_RDMA_WITH_IMM
858 IB_WC_WITH_IMM = (1<<1),
859 IB_WC_WITH_INVALIDATE = (1<<2),
860 IB_WC_IP_CSUM_OK = (1<<3),
861 IB_WC_WITH_SMAC = (1<<4),
862 IB_WC_WITH_VLAN = (1<<5),
863 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
869 struct ib_cqe *wr_cqe;
871 enum ib_wc_status status;
872 enum ib_wc_opcode opcode;
886 u8 port_num; /* valid only for DR SMPs on switches */
892 enum ib_cq_notify_flags {
893 IB_CQ_SOLICITED = 1 << 0,
894 IB_CQ_NEXT_COMP = 1 << 1,
895 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
896 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
904 enum ib_srq_attr_mask {
905 IB_SRQ_MAX_WR = 1 << 0,
906 IB_SRQ_LIMIT = 1 << 1,
915 struct ib_srq_init_attr {
916 void (*event_handler)(struct ib_event *, void *);
918 struct ib_srq_attr attr;
919 enum ib_srq_type srq_type;
923 struct ib_xrcd *xrcd;
944 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
945 * here (and in that order) since the MAD layer uses them as
946 * indices into a 2-entry table.
955 IB_QPT_RAW_ETHERTYPE,
956 IB_QPT_RAW_PACKET = 8,
960 /* Reserve a range for qp types internal to the low level driver.
961 * These qp types will not be visible at the IB core layer, so the
962 * IB_QPT_MAX usages should not be affected in the core layer
964 IB_QPT_RESERVED1 = 0x1000,
976 enum ib_qp_create_flags {
977 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
978 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
979 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
980 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
981 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
982 IB_QP_CREATE_NETIF_QP = 1 << 5,
983 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
984 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
985 /* reserve bits 26-31 for low level drivers' internal use */
986 IB_QP_CREATE_RESERVED_START = 1 << 26,
987 IB_QP_CREATE_RESERVED_END = 1 << 31,
991 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
992 * callback to destroy the passed in QP.
995 struct ib_qp_init_attr {
996 void (*event_handler)(struct ib_event *, void *);
998 struct ib_cq *send_cq;
999 struct ib_cq *recv_cq;
1001 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1002 struct ib_qp_cap cap;
1003 enum ib_sig_type sq_sig_type;
1004 enum ib_qp_type qp_type;
1005 enum ib_qp_create_flags create_flags;
1006 u8 port_num; /* special QP types only */
1009 struct ib_qp_open_attr {
1010 void (*event_handler)(struct ib_event *, void *);
1013 enum ib_qp_type qp_type;
1016 enum ib_rnr_timeout {
1017 IB_RNR_TIMER_655_36 = 0,
1018 IB_RNR_TIMER_000_01 = 1,
1019 IB_RNR_TIMER_000_02 = 2,
1020 IB_RNR_TIMER_000_03 = 3,
1021 IB_RNR_TIMER_000_04 = 4,
1022 IB_RNR_TIMER_000_06 = 5,
1023 IB_RNR_TIMER_000_08 = 6,
1024 IB_RNR_TIMER_000_12 = 7,
1025 IB_RNR_TIMER_000_16 = 8,
1026 IB_RNR_TIMER_000_24 = 9,
1027 IB_RNR_TIMER_000_32 = 10,
1028 IB_RNR_TIMER_000_48 = 11,
1029 IB_RNR_TIMER_000_64 = 12,
1030 IB_RNR_TIMER_000_96 = 13,
1031 IB_RNR_TIMER_001_28 = 14,
1032 IB_RNR_TIMER_001_92 = 15,
1033 IB_RNR_TIMER_002_56 = 16,
1034 IB_RNR_TIMER_003_84 = 17,
1035 IB_RNR_TIMER_005_12 = 18,
1036 IB_RNR_TIMER_007_68 = 19,
1037 IB_RNR_TIMER_010_24 = 20,
1038 IB_RNR_TIMER_015_36 = 21,
1039 IB_RNR_TIMER_020_48 = 22,
1040 IB_RNR_TIMER_030_72 = 23,
1041 IB_RNR_TIMER_040_96 = 24,
1042 IB_RNR_TIMER_061_44 = 25,
1043 IB_RNR_TIMER_081_92 = 26,
1044 IB_RNR_TIMER_122_88 = 27,
1045 IB_RNR_TIMER_163_84 = 28,
1046 IB_RNR_TIMER_245_76 = 29,
1047 IB_RNR_TIMER_327_68 = 30,
1048 IB_RNR_TIMER_491_52 = 31
1051 enum ib_qp_attr_mask {
1053 IB_QP_CUR_STATE = (1<<1),
1054 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1055 IB_QP_ACCESS_FLAGS = (1<<3),
1056 IB_QP_PKEY_INDEX = (1<<4),
1057 IB_QP_PORT = (1<<5),
1058 IB_QP_QKEY = (1<<6),
1060 IB_QP_PATH_MTU = (1<<8),
1061 IB_QP_TIMEOUT = (1<<9),
1062 IB_QP_RETRY_CNT = (1<<10),
1063 IB_QP_RNR_RETRY = (1<<11),
1064 IB_QP_RQ_PSN = (1<<12),
1065 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1066 IB_QP_ALT_PATH = (1<<14),
1067 IB_QP_MIN_RNR_TIMER = (1<<15),
1068 IB_QP_SQ_PSN = (1<<16),
1069 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1070 IB_QP_PATH_MIG_STATE = (1<<18),
1071 IB_QP_CAP = (1<<19),
1072 IB_QP_DEST_QPN = (1<<20),
1073 IB_QP_RESERVED1 = (1<<21),
1074 IB_QP_RESERVED2 = (1<<22),
1075 IB_QP_RESERVED3 = (1<<23),
1076 IB_QP_RESERVED4 = (1<<24),
1101 enum ib_qp_state qp_state;
1102 enum ib_qp_state cur_qp_state;
1103 enum ib_mtu path_mtu;
1104 enum ib_mig_state path_mig_state;
1109 int qp_access_flags;
1110 struct ib_qp_cap cap;
1111 struct ib_ah_attr ah_attr;
1112 struct ib_ah_attr alt_ah_attr;
1115 u8 en_sqd_async_notify;
1118 u8 max_dest_rd_atomic;
1130 IB_WR_RDMA_WRITE_WITH_IMM,
1132 IB_WR_SEND_WITH_IMM,
1134 IB_WR_ATOMIC_CMP_AND_SWP,
1135 IB_WR_ATOMIC_FETCH_AND_ADD,
1137 IB_WR_SEND_WITH_INV,
1138 IB_WR_RDMA_READ_WITH_INV,
1141 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1142 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1144 /* reserve values for low level drivers' internal use.
1145 * These values will not be used at all in the ib core layer.
1147 IB_WR_RESERVED1 = 0xf0,
1159 enum ib_send_flags {
1161 IB_SEND_SIGNALED = (1<<1),
1162 IB_SEND_SOLICITED = (1<<2),
1163 IB_SEND_INLINE = (1<<3),
1164 IB_SEND_IP_CSUM = (1<<4),
1166 /* reserve bits 26-31 for low level drivers' internal use */
1167 IB_SEND_RESERVED_START = (1 << 26),
1168 IB_SEND_RESERVED_END = (1 << 31),
1178 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1182 struct ib_send_wr *next;
1185 struct ib_cqe *wr_cqe;
1187 struct ib_sge *sg_list;
1189 enum ib_wr_opcode opcode;
1193 u32 invalidate_rkey;
1198 struct ib_send_wr wr;
1203 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1205 return container_of(wr, struct ib_rdma_wr, wr);
1208 struct ib_atomic_wr {
1209 struct ib_send_wr wr;
1213 u64 compare_add_mask;
1218 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1220 return container_of(wr, struct ib_atomic_wr, wr);
1224 struct ib_send_wr wr;
1231 u16 pkey_index; /* valid for GSI only */
1232 u8 port_num; /* valid for DR SMPs on switch only */
1235 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1237 return container_of(wr, struct ib_ud_wr, wr);
1241 struct ib_send_wr wr;
1247 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1249 return container_of(wr, struct ib_reg_wr, wr);
1252 struct ib_sig_handover_wr {
1253 struct ib_send_wr wr;
1254 struct ib_sig_attrs *sig_attrs;
1255 struct ib_mr *sig_mr;
1257 struct ib_sge *prot;
1260 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1262 return container_of(wr, struct ib_sig_handover_wr, wr);
1266 struct ib_recv_wr *next;
1269 struct ib_cqe *wr_cqe;
1271 struct ib_sge *sg_list;
1275 enum ib_access_flags {
1276 IB_ACCESS_LOCAL_WRITE = 1,
1277 IB_ACCESS_REMOTE_WRITE = (1<<1),
1278 IB_ACCESS_REMOTE_READ = (1<<2),
1279 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1280 IB_ACCESS_MW_BIND = (1<<4),
1281 IB_ZERO_BASED = (1<<5),
1282 IB_ACCESS_ON_DEMAND = (1<<6),
1286 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1287 * are hidden here instead of a uapi header!
1289 enum ib_mr_rereg_flags {
1290 IB_MR_REREG_TRANS = 1,
1291 IB_MR_REREG_PD = (1<<1),
1292 IB_MR_REREG_ACCESS = (1<<2),
1293 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1296 struct ib_fmr_attr {
1304 struct ib_ucontext {
1305 struct ib_device *device;
1306 struct list_head pd_list;
1307 struct list_head mr_list;
1308 struct list_head mw_list;
1309 struct list_head cq_list;
1310 struct list_head qp_list;
1311 struct list_head srq_list;
1312 struct list_head ah_list;
1313 struct list_head xrcd_list;
1314 struct list_head rule_list;
1318 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1319 struct rb_root umem_tree;
1321 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1322 * mmu notifiers registration.
1324 struct rw_semaphore umem_rwsem;
1325 void (*invalidate_range)(struct ib_umem *umem,
1326 unsigned long start, unsigned long end);
1328 struct mmu_notifier mn;
1329 atomic_t notifier_count;
1330 /* A list of umems that don't have private mmu notifier counters yet. */
1331 struct list_head no_private_counters;
1337 u64 user_handle; /* handle given to us by userspace */
1338 struct ib_ucontext *context; /* associated user context */
1339 void *object; /* containing object */
1340 struct list_head list; /* link to context's list */
1341 int id; /* index into kernel idr */
1343 struct rw_semaphore mutex; /* protects .live */
1344 struct rcu_head rcu; /* kfree_rcu() overhead */
1349 const void __user *inbuf;
1350 void __user *outbuf;
1357 struct ib_device *device;
1358 struct ib_uobject *uobject;
1359 atomic_t usecnt; /* count all resources */
1360 struct ib_mr *local_mr;
1364 struct ib_device *device;
1365 atomic_t usecnt; /* count all exposed resources */
1366 struct inode *inode;
1368 struct mutex tgt_qp_mutex;
1369 struct list_head tgt_qp_list;
1373 struct ib_device *device;
1375 struct ib_uobject *uobject;
1378 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1380 enum ib_poll_context {
1381 IB_POLL_DIRECT, /* caller context, no hw completions */
1382 IB_POLL_SOFTIRQ, /* poll from softirq context */
1383 IB_POLL_WORKQUEUE, /* poll from workqueue */
1387 struct ib_device *device;
1388 struct ib_uobject *uobject;
1389 ib_comp_handler comp_handler;
1390 void (*event_handler)(struct ib_event *, void *);
1393 atomic_t usecnt; /* count number of work queues */
1394 enum ib_poll_context poll_ctx;
1397 struct irq_poll iop;
1398 struct work_struct work;
1403 struct ib_device *device;
1405 struct ib_uobject *uobject;
1406 void (*event_handler)(struct ib_event *, void *);
1408 enum ib_srq_type srq_type;
1413 struct ib_xrcd *xrcd;
1421 struct ib_device *device;
1423 struct ib_cq *send_cq;
1424 struct ib_cq *recv_cq;
1426 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1427 struct list_head xrcd_list;
1428 /* count times opened, mcast attaches, flow attaches */
1430 struct list_head open_list;
1431 struct ib_qp *real_qp;
1432 struct ib_uobject *uobject;
1433 void (*event_handler)(struct ib_event *, void *);
1436 enum ib_qp_type qp_type;
1440 struct ib_device *device;
1442 struct ib_uobject *uobject;
1447 unsigned int page_size;
1451 struct ib_device *device;
1453 struct ib_uobject *uobject;
1455 enum ib_mw_type type;
1459 struct ib_device *device;
1461 struct list_head list;
1466 /* Supported steering options */
1467 enum ib_flow_attr_type {
1468 /* steering according to rule specifications */
1469 IB_FLOW_ATTR_NORMAL = 0x0,
1470 /* default unicast and multicast rule -
1471 * receive all Eth traffic which isn't steered to any QP
1473 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1474 /* default multicast rule -
1475 * receive all Eth multicast traffic which isn't steered to any QP
1477 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1478 /* sniffer rule - receive all port traffic */
1479 IB_FLOW_ATTR_SNIFFER = 0x3
1482 /* Supported steering header types */
1483 enum ib_flow_spec_type {
1485 IB_FLOW_SPEC_ETH = 0x20,
1486 IB_FLOW_SPEC_IB = 0x22,
1488 IB_FLOW_SPEC_IPV4 = 0x30,
1490 IB_FLOW_SPEC_TCP = 0x40,
1491 IB_FLOW_SPEC_UDP = 0x41
1493 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1494 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1496 /* Flow steering rule priority is set according to it's domain.
1497 * Lower domain value means higher priority.
1499 enum ib_flow_domain {
1500 IB_FLOW_DOMAIN_USER,
1501 IB_FLOW_DOMAIN_ETHTOOL,
1504 IB_FLOW_DOMAIN_NUM /* Must be last */
1507 enum ib_flow_flags {
1508 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1509 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1512 struct ib_flow_eth_filter {
1519 struct ib_flow_spec_eth {
1520 enum ib_flow_spec_type type;
1522 struct ib_flow_eth_filter val;
1523 struct ib_flow_eth_filter mask;
1526 struct ib_flow_ib_filter {
1531 struct ib_flow_spec_ib {
1532 enum ib_flow_spec_type type;
1534 struct ib_flow_ib_filter val;
1535 struct ib_flow_ib_filter mask;
1538 struct ib_flow_ipv4_filter {
1543 struct ib_flow_spec_ipv4 {
1544 enum ib_flow_spec_type type;
1546 struct ib_flow_ipv4_filter val;
1547 struct ib_flow_ipv4_filter mask;
1550 struct ib_flow_tcp_udp_filter {
1555 struct ib_flow_spec_tcp_udp {
1556 enum ib_flow_spec_type type;
1558 struct ib_flow_tcp_udp_filter val;
1559 struct ib_flow_tcp_udp_filter mask;
1562 union ib_flow_spec {
1564 enum ib_flow_spec_type type;
1567 struct ib_flow_spec_eth eth;
1568 struct ib_flow_spec_ib ib;
1569 struct ib_flow_spec_ipv4 ipv4;
1570 struct ib_flow_spec_tcp_udp tcp_udp;
1573 struct ib_flow_attr {
1574 enum ib_flow_attr_type type;
1580 /* Following are the optional layers according to user request
1581 * struct ib_flow_spec_xxx
1582 * struct ib_flow_spec_yyy
1588 struct ib_uobject *uobject;
1594 enum ib_process_mad_flags {
1595 IB_MAD_IGNORE_MKEY = 1,
1596 IB_MAD_IGNORE_BKEY = 2,
1597 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1600 enum ib_mad_result {
1601 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1602 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1603 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1604 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1607 #define IB_DEVICE_NAME_MAX 64
1611 struct ib_event_handler event_handler;
1612 struct ib_pkey_cache **pkey_cache;
1613 struct ib_gid_table **gid_cache;
1617 struct ib_dma_mapping_ops {
1618 int (*mapping_error)(struct ib_device *dev,
1620 u64 (*map_single)(struct ib_device *dev,
1621 void *ptr, size_t size,
1622 enum dma_data_direction direction);
1623 void (*unmap_single)(struct ib_device *dev,
1624 u64 addr, size_t size,
1625 enum dma_data_direction direction);
1626 u64 (*map_page)(struct ib_device *dev,
1627 struct page *page, unsigned long offset,
1629 enum dma_data_direction direction);
1630 void (*unmap_page)(struct ib_device *dev,
1631 u64 addr, size_t size,
1632 enum dma_data_direction direction);
1633 int (*map_sg)(struct ib_device *dev,
1634 struct scatterlist *sg, int nents,
1635 enum dma_data_direction direction);
1636 void (*unmap_sg)(struct ib_device *dev,
1637 struct scatterlist *sg, int nents,
1638 enum dma_data_direction direction);
1639 void (*sync_single_for_cpu)(struct ib_device *dev,
1642 enum dma_data_direction dir);
1643 void (*sync_single_for_device)(struct ib_device *dev,
1646 enum dma_data_direction dir);
1647 void *(*alloc_coherent)(struct ib_device *dev,
1651 void (*free_coherent)(struct ib_device *dev,
1652 size_t size, void *cpu_addr,
1658 struct ib_port_immutable {
1666 struct device *dma_device;
1668 char name[IB_DEVICE_NAME_MAX];
1670 struct list_head event_handler_list;
1671 spinlock_t event_handler_lock;
1673 spinlock_t client_data_lock;
1674 struct list_head core_list;
1675 /* Access to the client_data_list is protected by the client_data_lock
1676 * spinlock and the lists_rwsem read-write semaphore */
1677 struct list_head client_data_list;
1679 struct ib_cache cache;
1681 * port_immutable is indexed by port number
1683 struct ib_port_immutable *port_immutable;
1685 int num_comp_vectors;
1687 struct iw_cm_verbs *iwcm;
1689 int (*get_protocol_stats)(struct ib_device *device,
1690 union rdma_protocol_stats *stats);
1691 int (*query_device)(struct ib_device *device,
1692 struct ib_device_attr *device_attr,
1693 struct ib_udata *udata);
1694 int (*query_port)(struct ib_device *device,
1696 struct ib_port_attr *port_attr);
1697 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1699 /* When calling get_netdev, the HW vendor's driver should return the
1700 * net device of device @device at port @port_num or NULL if such
1701 * a net device doesn't exist. The vendor driver should call dev_hold
1702 * on this net device. The HW vendor's device driver must guarantee
1703 * that this function returns NULL before the net device reaches
1704 * NETDEV_UNREGISTER_FINAL state.
1706 struct net_device *(*get_netdev)(struct ib_device *device,
1708 int (*query_gid)(struct ib_device *device,
1709 u8 port_num, int index,
1711 /* When calling add_gid, the HW vendor's driver should
1712 * add the gid of device @device at gid index @index of
1713 * port @port_num to be @gid. Meta-info of that gid (for example,
1714 * the network device related to this gid is available
1715 * at @attr. @context allows the HW vendor driver to store extra
1716 * information together with a GID entry. The HW vendor may allocate
1717 * memory to contain this information and store it in @context when a
1718 * new GID entry is written to. Params are consistent until the next
1719 * call of add_gid or delete_gid. The function should return 0 on
1720 * success or error otherwise. The function could be called
1721 * concurrently for different ports. This function is only called
1722 * when roce_gid_table is used.
1724 int (*add_gid)(struct ib_device *device,
1727 const union ib_gid *gid,
1728 const struct ib_gid_attr *attr,
1730 /* When calling del_gid, the HW vendor's driver should delete the
1731 * gid of device @device at gid index @index of port @port_num.
1732 * Upon the deletion of a GID entry, the HW vendor must free any
1733 * allocated memory. The caller will clear @context afterwards.
1734 * This function is only called when roce_gid_table is used.
1736 int (*del_gid)(struct ib_device *device,
1740 int (*query_pkey)(struct ib_device *device,
1741 u8 port_num, u16 index, u16 *pkey);
1742 int (*modify_device)(struct ib_device *device,
1743 int device_modify_mask,
1744 struct ib_device_modify *device_modify);
1745 int (*modify_port)(struct ib_device *device,
1746 u8 port_num, int port_modify_mask,
1747 struct ib_port_modify *port_modify);
1748 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1749 struct ib_udata *udata);
1750 int (*dealloc_ucontext)(struct ib_ucontext *context);
1751 int (*mmap)(struct ib_ucontext *context,
1752 struct vm_area_struct *vma);
1753 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1754 struct ib_ucontext *context,
1755 struct ib_udata *udata);
1756 int (*dealloc_pd)(struct ib_pd *pd);
1757 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1758 struct ib_ah_attr *ah_attr);
1759 int (*modify_ah)(struct ib_ah *ah,
1760 struct ib_ah_attr *ah_attr);
1761 int (*query_ah)(struct ib_ah *ah,
1762 struct ib_ah_attr *ah_attr);
1763 int (*destroy_ah)(struct ib_ah *ah);
1764 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1765 struct ib_srq_init_attr *srq_init_attr,
1766 struct ib_udata *udata);
1767 int (*modify_srq)(struct ib_srq *srq,
1768 struct ib_srq_attr *srq_attr,
1769 enum ib_srq_attr_mask srq_attr_mask,
1770 struct ib_udata *udata);
1771 int (*query_srq)(struct ib_srq *srq,
1772 struct ib_srq_attr *srq_attr);
1773 int (*destroy_srq)(struct ib_srq *srq);
1774 int (*post_srq_recv)(struct ib_srq *srq,
1775 struct ib_recv_wr *recv_wr,
1776 struct ib_recv_wr **bad_recv_wr);
1777 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1778 struct ib_qp_init_attr *qp_init_attr,
1779 struct ib_udata *udata);
1780 int (*modify_qp)(struct ib_qp *qp,
1781 struct ib_qp_attr *qp_attr,
1783 struct ib_udata *udata);
1784 int (*query_qp)(struct ib_qp *qp,
1785 struct ib_qp_attr *qp_attr,
1787 struct ib_qp_init_attr *qp_init_attr);
1788 int (*destroy_qp)(struct ib_qp *qp);
1789 int (*post_send)(struct ib_qp *qp,
1790 struct ib_send_wr *send_wr,
1791 struct ib_send_wr **bad_send_wr);
1792 int (*post_recv)(struct ib_qp *qp,
1793 struct ib_recv_wr *recv_wr,
1794 struct ib_recv_wr **bad_recv_wr);
1795 struct ib_cq * (*create_cq)(struct ib_device *device,
1796 const struct ib_cq_init_attr *attr,
1797 struct ib_ucontext *context,
1798 struct ib_udata *udata);
1799 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1801 int (*destroy_cq)(struct ib_cq *cq);
1802 int (*resize_cq)(struct ib_cq *cq, int cqe,
1803 struct ib_udata *udata);
1804 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1806 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1807 int (*req_notify_cq)(struct ib_cq *cq,
1808 enum ib_cq_notify_flags flags);
1809 int (*req_ncomp_notif)(struct ib_cq *cq,
1811 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1812 int mr_access_flags);
1813 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1814 u64 start, u64 length,
1816 int mr_access_flags,
1817 struct ib_udata *udata);
1818 int (*rereg_user_mr)(struct ib_mr *mr,
1820 u64 start, u64 length,
1822 int mr_access_flags,
1824 struct ib_udata *udata);
1825 int (*dereg_mr)(struct ib_mr *mr);
1826 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1827 enum ib_mr_type mr_type,
1829 int (*map_mr_sg)(struct ib_mr *mr,
1830 struct scatterlist *sg,
1832 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1833 enum ib_mw_type type,
1834 struct ib_udata *udata);
1835 int (*dealloc_mw)(struct ib_mw *mw);
1836 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1837 int mr_access_flags,
1838 struct ib_fmr_attr *fmr_attr);
1839 int (*map_phys_fmr)(struct ib_fmr *fmr,
1840 u64 *page_list, int list_len,
1842 int (*unmap_fmr)(struct list_head *fmr_list);
1843 int (*dealloc_fmr)(struct ib_fmr *fmr);
1844 int (*attach_mcast)(struct ib_qp *qp,
1847 int (*detach_mcast)(struct ib_qp *qp,
1850 int (*process_mad)(struct ib_device *device,
1851 int process_mad_flags,
1853 const struct ib_wc *in_wc,
1854 const struct ib_grh *in_grh,
1855 const struct ib_mad_hdr *in_mad,
1857 struct ib_mad_hdr *out_mad,
1858 size_t *out_mad_size,
1859 u16 *out_mad_pkey_index);
1860 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1861 struct ib_ucontext *ucontext,
1862 struct ib_udata *udata);
1863 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1864 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1868 int (*destroy_flow)(struct ib_flow *flow_id);
1869 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1870 struct ib_mr_status *mr_status);
1871 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1872 void (*drain_rq)(struct ib_qp *qp);
1873 void (*drain_sq)(struct ib_qp *qp);
1874 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1876 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1877 struct ifla_vf_info *ivf);
1878 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1879 struct ifla_vf_stats *stats);
1880 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1883 struct ib_dma_mapping_ops *dma_ops;
1885 struct module *owner;
1887 struct kobject *ports_parent;
1888 struct list_head port_list;
1891 IB_DEV_UNINITIALIZED,
1897 u64 uverbs_cmd_mask;
1898 u64 uverbs_ex_cmd_mask;
1906 struct ib_device_attr attrs;
1909 * The following mandatory functions are used only at device
1910 * registration. Keep functions such as these at the end of this
1911 * structure to avoid cache line misses when accessing struct ib_device
1914 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1919 void (*add) (struct ib_device *);
1920 void (*remove)(struct ib_device *, void *client_data);
1922 /* Returns the net_dev belonging to this ib_client and matching the
1924 * @dev: An RDMA device that the net_dev use for communication.
1925 * @port: A physical port number on the RDMA device.
1926 * @pkey: P_Key that the net_dev uses if applicable.
1927 * @gid: A GID that the net_dev uses to communicate.
1928 * @addr: An IP address the net_dev is configured with.
1929 * @client_data: The device's client data set by ib_set_client_data().
1931 * An ib_client that implements a net_dev on top of RDMA devices
1932 * (such as IP over IB) should implement this callback, allowing the
1933 * rdma_cm module to find the right net_dev for a given request.
1935 * The caller is responsible for calling dev_put on the returned
1937 struct net_device *(*get_net_dev_by_params)(
1938 struct ib_device *dev,
1941 const union ib_gid *gid,
1942 const struct sockaddr *addr,
1944 struct list_head list;
1947 struct ib_device *ib_alloc_device(size_t size);
1948 void ib_dealloc_device(struct ib_device *device);
1950 int ib_register_device(struct ib_device *device,
1951 int (*port_callback)(struct ib_device *,
1952 u8, struct kobject *));
1953 void ib_unregister_device(struct ib_device *device);
1955 int ib_register_client (struct ib_client *client);
1956 void ib_unregister_client(struct ib_client *client);
1958 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1959 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1962 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1964 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1967 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1969 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1972 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
1976 const void __user *p = udata->inbuf + offset;
1980 if (len > USHRT_MAX)
1983 buf = kmalloc(len, GFP_KERNEL);
1987 if (copy_from_user(buf, p, len))
1990 ret = !memchr_inv(buf, 0, len);
1998 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1999 * contains all required attributes and no attributes not allowed for
2000 * the given QP state transition.
2001 * @cur_state: Current QP state
2002 * @next_state: Next QP state
2004 * @mask: Mask of supplied QP attributes
2005 * @ll : link layer of port
2007 * This function is a helper function that a low-level driver's
2008 * modify_qp method can use to validate the consumer's input. It
2009 * checks that cur_state and next_state are valid QP states, that a
2010 * transition from cur_state to next_state is allowed by the IB spec,
2011 * and that the attribute mask supplied is allowed for the transition.
2013 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2014 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2015 enum rdma_link_layer ll);
2017 int ib_register_event_handler (struct ib_event_handler *event_handler);
2018 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2019 void ib_dispatch_event(struct ib_event *event);
2021 int ib_query_port(struct ib_device *device,
2022 u8 port_num, struct ib_port_attr *port_attr);
2024 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2028 * rdma_cap_ib_switch - Check if the device is IB switch
2029 * @device: Device to check
2031 * Device driver is responsible for setting is_switch bit on
2032 * in ib_device structure at init time.
2034 * Return: true if the device is IB switch.
2036 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2038 return device->is_switch;
2042 * rdma_start_port - Return the first valid port number for the device
2045 * @device: Device to be checked
2047 * Return start port number
2049 static inline u8 rdma_start_port(const struct ib_device *device)
2051 return rdma_cap_ib_switch(device) ? 0 : 1;
2055 * rdma_end_port - Return the last valid port number for the device
2058 * @device: Device to be checked
2060 * Return last port number
2062 static inline u8 rdma_end_port(const struct ib_device *device)
2064 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2067 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2069 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2072 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2074 return device->port_immutable[port_num].core_cap_flags &
2075 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2078 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2080 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2083 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2085 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2088 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2090 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2093 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2095 return rdma_protocol_ib(device, port_num) ||
2096 rdma_protocol_roce(device, port_num);
2100 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2101 * Management Datagrams.
2102 * @device: Device to check
2103 * @port_num: Port number to check
2105 * Management Datagrams (MAD) are a required part of the InfiniBand
2106 * specification and are supported on all InfiniBand devices. A slightly
2107 * extended version are also supported on OPA interfaces.
2109 * Return: true if the port supports sending/receiving of MAD packets.
2111 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2113 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2117 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2118 * Management Datagrams.
2119 * @device: Device to check
2120 * @port_num: Port number to check
2122 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2123 * datagrams with their own versions. These OPA MADs share many but not all of
2124 * the characteristics of InfiniBand MADs.
2126 * OPA MADs differ in the following ways:
2128 * 1) MADs are variable size up to 2K
2129 * IBTA defined MADs remain fixed at 256 bytes
2130 * 2) OPA SMPs must carry valid PKeys
2131 * 3) OPA SMP packets are a different format
2133 * Return: true if the port supports OPA MAD packet formats.
2135 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2137 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2138 == RDMA_CORE_CAP_OPA_MAD;
2142 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2143 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2144 * @device: Device to check
2145 * @port_num: Port number to check
2147 * Each InfiniBand node is required to provide a Subnet Management Agent
2148 * that the subnet manager can access. Prior to the fabric being fully
2149 * configured by the subnet manager, the SMA is accessed via a well known
2150 * interface called the Subnet Management Interface (SMI). This interface
2151 * uses directed route packets to communicate with the SM to get around the
2152 * chicken and egg problem of the SM needing to know what's on the fabric
2153 * in order to configure the fabric, and needing to configure the fabric in
2154 * order to send packets to the devices on the fabric. These directed
2155 * route packets do not need the fabric fully configured in order to reach
2156 * their destination. The SMI is the only method allowed to send
2157 * directed route packets on an InfiniBand fabric.
2159 * Return: true if the port provides an SMI.
2161 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2163 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2167 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2168 * Communication Manager.
2169 * @device: Device to check
2170 * @port_num: Port number to check
2172 * The InfiniBand Communication Manager is one of many pre-defined General
2173 * Service Agents (GSA) that are accessed via the General Service
2174 * Interface (GSI). It's role is to facilitate establishment of connections
2175 * between nodes as well as other management related tasks for established
2178 * Return: true if the port supports an IB CM (this does not guarantee that
2179 * a CM is actually running however).
2181 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2183 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2187 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2188 * Communication Manager.
2189 * @device: Device to check
2190 * @port_num: Port number to check
2192 * Similar to above, but specific to iWARP connections which have a different
2193 * managment protocol than InfiniBand.
2195 * Return: true if the port supports an iWARP CM (this does not guarantee that
2196 * a CM is actually running however).
2198 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2200 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2204 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2205 * Subnet Administration.
2206 * @device: Device to check
2207 * @port_num: Port number to check
2209 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2210 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2211 * fabrics, devices should resolve routes to other hosts by contacting the
2212 * SA to query the proper route.
2214 * Return: true if the port should act as a client to the fabric Subnet
2215 * Administration interface. This does not imply that the SA service is
2218 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2220 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2224 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2226 * @device: Device to check
2227 * @port_num: Port number to check
2229 * InfiniBand multicast registration is more complex than normal IPv4 or
2230 * IPv6 multicast registration. Each Host Channel Adapter must register
2231 * with the Subnet Manager when it wishes to join a multicast group. It
2232 * should do so only once regardless of how many queue pairs it subscribes
2233 * to this group. And it should leave the group only after all queue pairs
2234 * attached to the group have been detached.
2236 * Return: true if the port must undertake the additional adminstrative
2237 * overhead of registering/unregistering with the SM and tracking of the
2238 * total number of queue pairs attached to the multicast group.
2240 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2242 return rdma_cap_ib_sa(device, port_num);
2246 * rdma_cap_af_ib - Check if the port of device has the capability
2247 * Native Infiniband Address.
2248 * @device: Device to check
2249 * @port_num: Port number to check
2251 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2252 * GID. RoCE uses a different mechanism, but still generates a GID via
2253 * a prescribed mechanism and port specific data.
2255 * Return: true if the port uses a GID address to identify devices on the
2258 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2260 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2264 * rdma_cap_eth_ah - Check if the port of device has the capability
2265 * Ethernet Address Handle.
2266 * @device: Device to check
2267 * @port_num: Port number to check
2269 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2270 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2271 * port. Normally, packet headers are generated by the sending host
2272 * adapter, but when sending connectionless datagrams, we must manually
2273 * inject the proper headers for the fabric we are communicating over.
2275 * Return: true if we are running as a RoCE port and must force the
2276 * addition of a Global Route Header built from our Ethernet Address
2277 * Handle into our header list for connectionless packets.
2279 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2281 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2285 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2288 * @port_num: Port number
2290 * This MAD size includes the MAD headers and MAD payload. No other headers
2293 * Return the max MAD size required by the Port. Will return 0 if the port
2294 * does not support MADs
2296 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2298 return device->port_immutable[port_num].max_mad_size;
2302 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2303 * @device: Device to check
2304 * @port_num: Port number to check
2306 * RoCE GID table mechanism manages the various GIDs for a device.
2308 * NOTE: if allocating the port's GID table has failed, this call will still
2309 * return true, but any RoCE GID table API will fail.
2311 * Return: true if the port uses RoCE GID table mechanism in order to manage
2314 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2317 return rdma_protocol_roce(device, port_num) &&
2318 device->add_gid && device->del_gid;
2321 int ib_query_gid(struct ib_device *device,
2322 u8 port_num, int index, union ib_gid *gid,
2323 struct ib_gid_attr *attr);
2325 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2327 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2328 struct ifla_vf_info *info);
2329 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2330 struct ifla_vf_stats *stats);
2331 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2334 int ib_query_pkey(struct ib_device *device,
2335 u8 port_num, u16 index, u16 *pkey);
2337 int ib_modify_device(struct ib_device *device,
2338 int device_modify_mask,
2339 struct ib_device_modify *device_modify);
2341 int ib_modify_port(struct ib_device *device,
2342 u8 port_num, int port_modify_mask,
2343 struct ib_port_modify *port_modify);
2345 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2346 enum ib_gid_type gid_type, struct net_device *ndev,
2347 u8 *port_num, u16 *index);
2349 int ib_find_pkey(struct ib_device *device,
2350 u8 port_num, u16 pkey, u16 *index);
2352 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2354 void ib_dealloc_pd(struct ib_pd *pd);
2357 * ib_create_ah - Creates an address handle for the given address vector.
2358 * @pd: The protection domain associated with the address handle.
2359 * @ah_attr: The attributes of the address vector.
2361 * The address handle is used to reference a local or global destination
2362 * in all UD QP post sends.
2364 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2367 * ib_init_ah_from_wc - Initializes address handle attributes from a
2369 * @device: Device on which the received message arrived.
2370 * @port_num: Port on which the received message arrived.
2371 * @wc: Work completion associated with the received message.
2372 * @grh: References the received global route header. This parameter is
2373 * ignored unless the work completion indicates that the GRH is valid.
2374 * @ah_attr: Returned attributes that can be used when creating an address
2375 * handle for replying to the message.
2377 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2378 const struct ib_wc *wc, const struct ib_grh *grh,
2379 struct ib_ah_attr *ah_attr);
2382 * ib_create_ah_from_wc - Creates an address handle associated with the
2383 * sender of the specified work completion.
2384 * @pd: The protection domain associated with the address handle.
2385 * @wc: Work completion information associated with a received message.
2386 * @grh: References the received global route header. This parameter is
2387 * ignored unless the work completion indicates that the GRH is valid.
2388 * @port_num: The outbound port number to associate with the address.
2390 * The address handle is used to reference a local or global destination
2391 * in all UD QP post sends.
2393 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2394 const struct ib_grh *grh, u8 port_num);
2397 * ib_modify_ah - Modifies the address vector associated with an address
2399 * @ah: The address handle to modify.
2400 * @ah_attr: The new address vector attributes to associate with the
2403 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2406 * ib_query_ah - Queries the address vector associated with an address
2408 * @ah: The address handle to query.
2409 * @ah_attr: The address vector attributes associated with the address
2412 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2415 * ib_destroy_ah - Destroys an address handle.
2416 * @ah: The address handle to destroy.
2418 int ib_destroy_ah(struct ib_ah *ah);
2421 * ib_create_srq - Creates a SRQ associated with the specified protection
2423 * @pd: The protection domain associated with the SRQ.
2424 * @srq_init_attr: A list of initial attributes required to create the
2425 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2426 * the actual capabilities of the created SRQ.
2428 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2429 * requested size of the SRQ, and set to the actual values allocated
2430 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2431 * will always be at least as large as the requested values.
2433 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2434 struct ib_srq_init_attr *srq_init_attr);
2437 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2438 * @srq: The SRQ to modify.
2439 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2440 * the current values of selected SRQ attributes are returned.
2441 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2442 * are being modified.
2444 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2445 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2446 * the number of receives queued drops below the limit.
2448 int ib_modify_srq(struct ib_srq *srq,
2449 struct ib_srq_attr *srq_attr,
2450 enum ib_srq_attr_mask srq_attr_mask);
2453 * ib_query_srq - Returns the attribute list and current values for the
2455 * @srq: The SRQ to query.
2456 * @srq_attr: The attributes of the specified SRQ.
2458 int ib_query_srq(struct ib_srq *srq,
2459 struct ib_srq_attr *srq_attr);
2462 * ib_destroy_srq - Destroys the specified SRQ.
2463 * @srq: The SRQ to destroy.
2465 int ib_destroy_srq(struct ib_srq *srq);
2468 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2469 * @srq: The SRQ to post the work request on.
2470 * @recv_wr: A list of work requests to post on the receive queue.
2471 * @bad_recv_wr: On an immediate failure, this parameter will reference
2472 * the work request that failed to be posted on the QP.
2474 static inline int ib_post_srq_recv(struct ib_srq *srq,
2475 struct ib_recv_wr *recv_wr,
2476 struct ib_recv_wr **bad_recv_wr)
2478 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2482 * ib_create_qp - Creates a QP associated with the specified protection
2484 * @pd: The protection domain associated with the QP.
2485 * @qp_init_attr: A list of initial attributes required to create the
2486 * QP. If QP creation succeeds, then the attributes are updated to
2487 * the actual capabilities of the created QP.
2489 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2490 struct ib_qp_init_attr *qp_init_attr);
2493 * ib_modify_qp - Modifies the attributes for the specified QP and then
2494 * transitions the QP to the given state.
2495 * @qp: The QP to modify.
2496 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2497 * the current values of selected QP attributes are returned.
2498 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2499 * are being modified.
2501 int ib_modify_qp(struct ib_qp *qp,
2502 struct ib_qp_attr *qp_attr,
2506 * ib_query_qp - Returns the attribute list and current values for the
2508 * @qp: The QP to query.
2509 * @qp_attr: The attributes of the specified QP.
2510 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2511 * @qp_init_attr: Additional attributes of the selected QP.
2513 * The qp_attr_mask may be used to limit the query to gathering only the
2514 * selected attributes.
2516 int ib_query_qp(struct ib_qp *qp,
2517 struct ib_qp_attr *qp_attr,
2519 struct ib_qp_init_attr *qp_init_attr);
2522 * ib_destroy_qp - Destroys the specified QP.
2523 * @qp: The QP to destroy.
2525 int ib_destroy_qp(struct ib_qp *qp);
2528 * ib_open_qp - Obtain a reference to an existing sharable QP.
2529 * @xrcd - XRC domain
2530 * @qp_open_attr: Attributes identifying the QP to open.
2532 * Returns a reference to a sharable QP.
2534 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2535 struct ib_qp_open_attr *qp_open_attr);
2538 * ib_close_qp - Release an external reference to a QP.
2539 * @qp: The QP handle to release
2541 * The opened QP handle is released by the caller. The underlying
2542 * shared QP is not destroyed until all internal references are released.
2544 int ib_close_qp(struct ib_qp *qp);
2547 * ib_post_send - Posts a list of work requests to the send queue of
2549 * @qp: The QP to post the work request on.
2550 * @send_wr: A list of work requests to post on the send queue.
2551 * @bad_send_wr: On an immediate failure, this parameter will reference
2552 * the work request that failed to be posted on the QP.
2554 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2555 * error is returned, the QP state shall not be affected,
2556 * ib_post_send() will return an immediate error after queueing any
2557 * earlier work requests in the list.
2559 static inline int ib_post_send(struct ib_qp *qp,
2560 struct ib_send_wr *send_wr,
2561 struct ib_send_wr **bad_send_wr)
2563 return qp->device->post_send(qp, send_wr, bad_send_wr);
2567 * ib_post_recv - Posts a list of work requests to the receive queue of
2569 * @qp: The QP to post the work request on.
2570 * @recv_wr: A list of work requests to post on the receive queue.
2571 * @bad_recv_wr: On an immediate failure, this parameter will reference
2572 * the work request that failed to be posted on the QP.
2574 static inline int ib_post_recv(struct ib_qp *qp,
2575 struct ib_recv_wr *recv_wr,
2576 struct ib_recv_wr **bad_recv_wr)
2578 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2581 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2582 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2583 void ib_free_cq(struct ib_cq *cq);
2584 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2587 * ib_create_cq - Creates a CQ on the specified device.
2588 * @device: The device on which to create the CQ.
2589 * @comp_handler: A user-specified callback that is invoked when a
2590 * completion event occurs on the CQ.
2591 * @event_handler: A user-specified callback that is invoked when an
2592 * asynchronous event not associated with a completion occurs on the CQ.
2593 * @cq_context: Context associated with the CQ returned to the user via
2594 * the associated completion and event handlers.
2595 * @cq_attr: The attributes the CQ should be created upon.
2597 * Users can examine the cq structure to determine the actual CQ size.
2599 struct ib_cq *ib_create_cq(struct ib_device *device,
2600 ib_comp_handler comp_handler,
2601 void (*event_handler)(struct ib_event *, void *),
2603 const struct ib_cq_init_attr *cq_attr);
2606 * ib_resize_cq - Modifies the capacity of the CQ.
2607 * @cq: The CQ to resize.
2608 * @cqe: The minimum size of the CQ.
2610 * Users can examine the cq structure to determine the actual CQ size.
2612 int ib_resize_cq(struct ib_cq *cq, int cqe);
2615 * ib_modify_cq - Modifies moderation params of the CQ
2616 * @cq: The CQ to modify.
2617 * @cq_count: number of CQEs that will trigger an event
2618 * @cq_period: max period of time in usec before triggering an event
2621 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2624 * ib_destroy_cq - Destroys the specified CQ.
2625 * @cq: The CQ to destroy.
2627 int ib_destroy_cq(struct ib_cq *cq);
2630 * ib_poll_cq - poll a CQ for completion(s)
2631 * @cq:the CQ being polled
2632 * @num_entries:maximum number of completions to return
2633 * @wc:array of at least @num_entries &struct ib_wc where completions
2636 * Poll a CQ for (possibly multiple) completions. If the return value
2637 * is < 0, an error occurred. If the return value is >= 0, it is the
2638 * number of completions returned. If the return value is
2639 * non-negative and < num_entries, then the CQ was emptied.
2641 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2644 return cq->device->poll_cq(cq, num_entries, wc);
2648 * ib_peek_cq - Returns the number of unreaped completions currently
2649 * on the specified CQ.
2650 * @cq: The CQ to peek.
2651 * @wc_cnt: A minimum number of unreaped completions to check for.
2653 * If the number of unreaped completions is greater than or equal to wc_cnt,
2654 * this function returns wc_cnt, otherwise, it returns the actual number of
2655 * unreaped completions.
2657 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2660 * ib_req_notify_cq - Request completion notification on a CQ.
2661 * @cq: The CQ to generate an event for.
2663 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2664 * to request an event on the next solicited event or next work
2665 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2666 * may also be |ed in to request a hint about missed events, as
2670 * < 0 means an error occurred while requesting notification
2671 * == 0 means notification was requested successfully, and if
2672 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2673 * were missed and it is safe to wait for another event. In
2674 * this case is it guaranteed that any work completions added
2675 * to the CQ since the last CQ poll will trigger a completion
2676 * notification event.
2677 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2678 * in. It means that the consumer must poll the CQ again to
2679 * make sure it is empty to avoid missing an event because of a
2680 * race between requesting notification and an entry being
2681 * added to the CQ. This return value means it is possible
2682 * (but not guaranteed) that a work completion has been added
2683 * to the CQ since the last poll without triggering a
2684 * completion notification event.
2686 static inline int ib_req_notify_cq(struct ib_cq *cq,
2687 enum ib_cq_notify_flags flags)
2689 return cq->device->req_notify_cq(cq, flags);
2693 * ib_req_ncomp_notif - Request completion notification when there are
2694 * at least the specified number of unreaped completions on the CQ.
2695 * @cq: The CQ to generate an event for.
2696 * @wc_cnt: The number of unreaped completions that should be on the
2697 * CQ before an event is generated.
2699 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2701 return cq->device->req_ncomp_notif ?
2702 cq->device->req_ncomp_notif(cq, wc_cnt) :
2707 * ib_get_dma_mr - Returns a memory region for system memory that is
2709 * @pd: The protection domain associated with the memory region.
2710 * @mr_access_flags: Specifies the memory access rights.
2712 * Note that the ib_dma_*() functions defined below must be used
2713 * to create/destroy addresses used with the Lkey or Rkey returned
2714 * by ib_get_dma_mr().
2716 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2719 * ib_dma_mapping_error - check a DMA addr for error
2720 * @dev: The device for which the dma_addr was created
2721 * @dma_addr: The DMA address to check
2723 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2726 return dev->dma_ops->mapping_error(dev, dma_addr);
2727 return dma_mapping_error(dev->dma_device, dma_addr);
2731 * ib_dma_map_single - Map a kernel virtual address to DMA address
2732 * @dev: The device for which the dma_addr is to be created
2733 * @cpu_addr: The kernel virtual address
2734 * @size: The size of the region in bytes
2735 * @direction: The direction of the DMA
2737 static inline u64 ib_dma_map_single(struct ib_device *dev,
2738 void *cpu_addr, size_t size,
2739 enum dma_data_direction direction)
2742 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2743 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2747 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2748 * @dev: The device for which the DMA address was created
2749 * @addr: The DMA address
2750 * @size: The size of the region in bytes
2751 * @direction: The direction of the DMA
2753 static inline void ib_dma_unmap_single(struct ib_device *dev,
2754 u64 addr, size_t size,
2755 enum dma_data_direction direction)
2758 dev->dma_ops->unmap_single(dev, addr, size, direction);
2760 dma_unmap_single(dev->dma_device, addr, size, direction);
2763 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2764 void *cpu_addr, size_t size,
2765 enum dma_data_direction direction,
2766 struct dma_attrs *attrs)
2768 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2772 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2773 u64 addr, size_t size,
2774 enum dma_data_direction direction,
2775 struct dma_attrs *attrs)
2777 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2782 * ib_dma_map_page - Map a physical page to DMA address
2783 * @dev: The device for which the dma_addr is to be created
2784 * @page: The page to be mapped
2785 * @offset: The offset within the page
2786 * @size: The size of the region in bytes
2787 * @direction: The direction of the DMA
2789 static inline u64 ib_dma_map_page(struct ib_device *dev,
2791 unsigned long offset,
2793 enum dma_data_direction direction)
2796 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2797 return dma_map_page(dev->dma_device, page, offset, size, direction);
2801 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2802 * @dev: The device for which the DMA address was created
2803 * @addr: The DMA address
2804 * @size: The size of the region in bytes
2805 * @direction: The direction of the DMA
2807 static inline void ib_dma_unmap_page(struct ib_device *dev,
2808 u64 addr, size_t size,
2809 enum dma_data_direction direction)
2812 dev->dma_ops->unmap_page(dev, addr, size, direction);
2814 dma_unmap_page(dev->dma_device, addr, size, direction);
2818 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2819 * @dev: The device for which the DMA addresses are to be created
2820 * @sg: The array of scatter/gather entries
2821 * @nents: The number of scatter/gather entries
2822 * @direction: The direction of the DMA
2824 static inline int ib_dma_map_sg(struct ib_device *dev,
2825 struct scatterlist *sg, int nents,
2826 enum dma_data_direction direction)
2829 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2830 return dma_map_sg(dev->dma_device, sg, nents, direction);
2834 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2835 * @dev: The device for which the DMA addresses were created
2836 * @sg: The array of scatter/gather entries
2837 * @nents: The number of scatter/gather entries
2838 * @direction: The direction of the DMA
2840 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2841 struct scatterlist *sg, int nents,
2842 enum dma_data_direction direction)
2845 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2847 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2850 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2851 struct scatterlist *sg, int nents,
2852 enum dma_data_direction direction,
2853 struct dma_attrs *attrs)
2855 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2858 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2859 struct scatterlist *sg, int nents,
2860 enum dma_data_direction direction,
2861 struct dma_attrs *attrs)
2863 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2866 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2867 * @dev: The device for which the DMA addresses were created
2868 * @sg: The scatter/gather entry
2870 * Note: this function is obsolete. To do: change all occurrences of
2871 * ib_sg_dma_address() into sg_dma_address().
2873 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2874 struct scatterlist *sg)
2876 return sg_dma_address(sg);
2880 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2881 * @dev: The device for which the DMA addresses were created
2882 * @sg: The scatter/gather entry
2884 * Note: this function is obsolete. To do: change all occurrences of
2885 * ib_sg_dma_len() into sg_dma_len().
2887 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2888 struct scatterlist *sg)
2890 return sg_dma_len(sg);
2894 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2895 * @dev: The device for which the DMA address was created
2896 * @addr: The DMA address
2897 * @size: The size of the region in bytes
2898 * @dir: The direction of the DMA
2900 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2903 enum dma_data_direction dir)
2906 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2908 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2912 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2913 * @dev: The device for which the DMA address was created
2914 * @addr: The DMA address
2915 * @size: The size of the region in bytes
2916 * @dir: The direction of the DMA
2918 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2921 enum dma_data_direction dir)
2924 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2926 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2930 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2931 * @dev: The device for which the DMA address is requested
2932 * @size: The size of the region to allocate in bytes
2933 * @dma_handle: A pointer for returning the DMA address of the region
2934 * @flag: memory allocator flags
2936 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2942 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2947 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2948 *dma_handle = handle;
2954 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2955 * @dev: The device for which the DMA addresses were allocated
2956 * @size: The size of the region
2957 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2958 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2960 static inline void ib_dma_free_coherent(struct ib_device *dev,
2961 size_t size, void *cpu_addr,
2965 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2967 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2971 * ib_dereg_mr - Deregisters a memory region and removes it from the
2972 * HCA translation table.
2973 * @mr: The memory region to deregister.
2975 * This function can fail, if the memory region has memory windows bound to it.
2977 int ib_dereg_mr(struct ib_mr *mr);
2979 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2980 enum ib_mr_type mr_type,
2984 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2986 * @mr - struct ib_mr pointer to be updated.
2987 * @newkey - new key to be used.
2989 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2991 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2992 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2996 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2997 * for calculating a new rkey for type 2 memory windows.
2998 * @rkey - the rkey to increment.
3000 static inline u32 ib_inc_rkey(u32 rkey)
3002 const u32 mask = 0x000000ff;
3003 return ((rkey + 1) & mask) | (rkey & ~mask);
3007 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3008 * @pd: The protection domain associated with the unmapped region.
3009 * @mr_access_flags: Specifies the memory access rights.
3010 * @fmr_attr: Attributes of the unmapped region.
3012 * A fast memory region must be mapped before it can be used as part of
3015 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3016 int mr_access_flags,
3017 struct ib_fmr_attr *fmr_attr);
3020 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3021 * @fmr: The fast memory region to associate with the pages.
3022 * @page_list: An array of physical pages to map to the fast memory region.
3023 * @list_len: The number of pages in page_list.
3024 * @iova: The I/O virtual address to use with the mapped region.
3026 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3027 u64 *page_list, int list_len,
3030 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3034 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3035 * @fmr_list: A linked list of fast memory regions to unmap.
3037 int ib_unmap_fmr(struct list_head *fmr_list);
3040 * ib_dealloc_fmr - Deallocates a fast memory region.
3041 * @fmr: The fast memory region to deallocate.
3043 int ib_dealloc_fmr(struct ib_fmr *fmr);
3046 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3047 * @qp: QP to attach to the multicast group. The QP must be type
3049 * @gid: Multicast group GID.
3050 * @lid: Multicast group LID in host byte order.
3052 * In order to send and receive multicast packets, subnet
3053 * administration must have created the multicast group and configured
3054 * the fabric appropriately. The port associated with the specified
3055 * QP must also be a member of the multicast group.
3057 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3060 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3061 * @qp: QP to detach from the multicast group.
3062 * @gid: Multicast group GID.
3063 * @lid: Multicast group LID in host byte order.
3065 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3068 * ib_alloc_xrcd - Allocates an XRC domain.
3069 * @device: The device on which to allocate the XRC domain.
3071 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3074 * ib_dealloc_xrcd - Deallocates an XRC domain.
3075 * @xrcd: The XRC domain to deallocate.
3077 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3079 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3080 struct ib_flow_attr *flow_attr, int domain);
3081 int ib_destroy_flow(struct ib_flow *flow_id);
3083 static inline int ib_check_mr_access(int flags)
3086 * Local write permission is required if remote write or
3087 * remote atomic permission is also requested.
3089 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3090 !(flags & IB_ACCESS_LOCAL_WRITE))
3097 * ib_check_mr_status: lightweight check of MR status.
3098 * This routine may provide status checks on a selected
3099 * ib_mr. first use is for signature status check.
3101 * @mr: A memory region.
3102 * @check_mask: Bitmask of which checks to perform from
3103 * ib_mr_status_check enumeration.
3104 * @mr_status: The container of relevant status checks.
3105 * failed checks will be indicated in the status bitmask
3106 * and the relevant info shall be in the error item.
3108 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3109 struct ib_mr_status *mr_status);
3111 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3112 u16 pkey, const union ib_gid *gid,
3113 const struct sockaddr *addr);
3115 int ib_map_mr_sg(struct ib_mr *mr,
3116 struct scatterlist *sg,
3118 unsigned int page_size);
3121 ib_map_mr_sg_zbva(struct ib_mr *mr,
3122 struct scatterlist *sg,
3124 unsigned int page_size)
3128 n = ib_map_mr_sg(mr, sg, sg_nents, page_size);
3134 int ib_sg_to_pages(struct ib_mr *mr,
3135 struct scatterlist *sgl,
3137 int (*set_page)(struct ib_mr *, u64));
3139 void ib_drain_rq(struct ib_qp *qp);
3140 void ib_drain_sq(struct ib_qp *qp);
3141 void ib_drain_qp(struct ib_qp *qp);
3142 #endif /* IB_VERBS_H */