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 = (1ULL << 31),
221 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
222 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
223 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 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;
265 /* Corresponding bit will be set if qp type from
266 * 'enum ib_qp_type' is supported, e.g.
267 * supported_qpts |= 1 << IB_QPT_UD
270 u32 max_rwq_indirection_tables;
271 u32 max_rwq_indirection_table_size;
274 enum ib_cq_creation_flags {
275 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
276 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
279 struct ib_cq_init_attr {
285 struct ib_device_attr {
287 __be64 sys_image_guid;
295 u64 device_cap_flags;
305 int max_qp_init_rd_atom;
306 int max_ee_init_rd_atom;
307 enum ib_atomic_cap atomic_cap;
308 enum ib_atomic_cap masked_atomic_cap;
315 int max_mcast_qp_attach;
316 int max_total_mcast_qp_attach;
323 unsigned int max_fast_reg_page_list_len;
325 u8 local_ca_ack_delay;
328 struct ib_odp_caps odp_caps;
329 uint64_t timestamp_mask;
330 uint64_t hca_core_clock; /* in KHZ */
331 struct ib_rss_caps rss_caps;
343 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
346 case IB_MTU_256: return 256;
347 case IB_MTU_512: return 512;
348 case IB_MTU_1024: return 1024;
349 case IB_MTU_2048: return 2048;
350 case IB_MTU_4096: return 4096;
361 IB_PORT_ACTIVE_DEFER = 5
364 enum ib_port_cap_flags {
366 IB_PORT_NOTICE_SUP = 1 << 2,
367 IB_PORT_TRAP_SUP = 1 << 3,
368 IB_PORT_OPT_IPD_SUP = 1 << 4,
369 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
370 IB_PORT_SL_MAP_SUP = 1 << 6,
371 IB_PORT_MKEY_NVRAM = 1 << 7,
372 IB_PORT_PKEY_NVRAM = 1 << 8,
373 IB_PORT_LED_INFO_SUP = 1 << 9,
374 IB_PORT_SM_DISABLED = 1 << 10,
375 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
376 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
377 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
378 IB_PORT_CM_SUP = 1 << 16,
379 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
380 IB_PORT_REINIT_SUP = 1 << 18,
381 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
382 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
383 IB_PORT_DR_NOTICE_SUP = 1 << 21,
384 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
385 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
386 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
387 IB_PORT_CLIENT_REG_SUP = 1 << 25,
388 IB_PORT_IP_BASED_GIDS = 1 << 26,
398 static inline int ib_width_enum_to_int(enum ib_port_width width)
401 case IB_WIDTH_1X: return 1;
402 case IB_WIDTH_4X: return 4;
403 case IB_WIDTH_8X: return 8;
404 case IB_WIDTH_12X: return 12;
419 * struct rdma_hw_stats
420 * @timestamp - Used by the core code to track when the last update was
421 * @lifespan - Used by the core code to determine how old the counters
422 * should be before being updated again. Stored in jiffies, defaults
423 * to 10 milliseconds, drivers can override the default be specifying
424 * their own value during their allocation routine.
425 * @name - Array of pointers to static names used for the counters in
427 * @num_counters - How many hardware counters there are. If name is
428 * shorter than this number, a kernel oops will result. Driver authors
429 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
430 * in their code to prevent this.
431 * @value - Array of u64 counters that are accessed by the sysfs code and
432 * filled in by the drivers get_stats routine
434 struct rdma_hw_stats {
435 unsigned long timestamp;
436 unsigned long lifespan;
437 const char * const *names;
442 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
444 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
446 * @names - Array of static const char *
447 * @num_counters - How many elements in array
448 * @lifespan - How many milliseconds between updates
450 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
451 const char * const *names, int num_counters,
452 unsigned long lifespan)
454 struct rdma_hw_stats *stats;
456 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
460 stats->names = names;
461 stats->num_counters = num_counters;
462 stats->lifespan = msecs_to_jiffies(lifespan);
468 /* Define bits for the various functionality this port needs to be supported by
471 /* Management 0x00000FFF */
472 #define RDMA_CORE_CAP_IB_MAD 0x00000001
473 #define RDMA_CORE_CAP_IB_SMI 0x00000002
474 #define RDMA_CORE_CAP_IB_CM 0x00000004
475 #define RDMA_CORE_CAP_IW_CM 0x00000008
476 #define RDMA_CORE_CAP_IB_SA 0x00000010
477 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
479 /* Address format 0x000FF000 */
480 #define RDMA_CORE_CAP_AF_IB 0x00001000
481 #define RDMA_CORE_CAP_ETH_AH 0x00002000
483 /* Protocol 0xFFF00000 */
484 #define RDMA_CORE_CAP_PROT_IB 0x00100000
485 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
486 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
487 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
489 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_SMI \
492 | RDMA_CORE_CAP_IB_CM \
493 | RDMA_CORE_CAP_IB_SA \
494 | RDMA_CORE_CAP_AF_IB)
495 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
496 | RDMA_CORE_CAP_IB_MAD \
497 | RDMA_CORE_CAP_IB_CM \
498 | RDMA_CORE_CAP_AF_IB \
499 | RDMA_CORE_CAP_ETH_AH)
500 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
501 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
502 | RDMA_CORE_CAP_IB_MAD \
503 | RDMA_CORE_CAP_IB_CM \
504 | RDMA_CORE_CAP_AF_IB \
505 | RDMA_CORE_CAP_ETH_AH)
506 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
507 | RDMA_CORE_CAP_IW_CM)
508 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
509 | RDMA_CORE_CAP_OPA_MAD)
511 struct ib_port_attr {
513 enum ib_port_state state;
515 enum ib_mtu active_mtu;
535 enum ib_device_modify_flags {
536 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
537 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
540 struct ib_device_modify {
545 enum ib_port_modify_flags {
546 IB_PORT_SHUTDOWN = 1,
547 IB_PORT_INIT_TYPE = (1<<2),
548 IB_PORT_RESET_QKEY_CNTR = (1<<3)
551 struct ib_port_modify {
552 u32 set_port_cap_mask;
553 u32 clr_port_cap_mask;
561 IB_EVENT_QP_ACCESS_ERR,
565 IB_EVENT_PATH_MIG_ERR,
566 IB_EVENT_DEVICE_FATAL,
567 IB_EVENT_PORT_ACTIVE,
570 IB_EVENT_PKEY_CHANGE,
573 IB_EVENT_SRQ_LIMIT_REACHED,
574 IB_EVENT_QP_LAST_WQE_REACHED,
575 IB_EVENT_CLIENT_REREGISTER,
580 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
583 struct ib_device *device;
591 enum ib_event_type event;
594 struct ib_event_handler {
595 struct ib_device *device;
596 void (*handler)(struct ib_event_handler *, struct ib_event *);
597 struct list_head list;
600 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
602 (_ptr)->device = _device; \
603 (_ptr)->handler = _handler; \
604 INIT_LIST_HEAD(&(_ptr)->list); \
607 struct ib_global_route {
616 __be32 version_tclass_flow;
624 union rdma_network_hdr {
627 /* The IB spec states that if it's IPv4, the header
628 * is located in the last 20 bytes of the header.
631 struct iphdr roce4grh;
636 IB_MULTICAST_QPN = 0xffffff
639 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
640 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
647 IB_RATE_PORT_CURRENT = 0,
648 IB_RATE_2_5_GBPS = 2,
656 IB_RATE_120_GBPS = 10,
657 IB_RATE_14_GBPS = 11,
658 IB_RATE_56_GBPS = 12,
659 IB_RATE_112_GBPS = 13,
660 IB_RATE_168_GBPS = 14,
661 IB_RATE_25_GBPS = 15,
662 IB_RATE_100_GBPS = 16,
663 IB_RATE_200_GBPS = 17,
664 IB_RATE_300_GBPS = 18
668 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
669 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
670 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
671 * @rate: rate to convert.
673 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
676 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
677 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
678 * @rate: rate to convert.
680 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
684 * enum ib_mr_type - memory region type
685 * @IB_MR_TYPE_MEM_REG: memory region that is used for
686 * normal registration
687 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
688 * signature operations (data-integrity
690 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
691 * register any arbitrary sg lists (without
692 * the normal mr constraints - see
697 IB_MR_TYPE_SIGNATURE,
703 * IB_SIG_TYPE_NONE: Unprotected.
704 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
706 enum ib_signature_type {
712 * Signature T10-DIF block-guard types
713 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
714 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
716 enum ib_t10_dif_bg_type {
722 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
724 * @bg_type: T10-DIF block guard type (CRC|CSUM)
725 * @pi_interval: protection information interval.
726 * @bg: seed of guard computation.
727 * @app_tag: application tag of guard block
728 * @ref_tag: initial guard block reference tag.
729 * @ref_remap: Indicate wethear the reftag increments each block
730 * @app_escape: Indicate to skip block check if apptag=0xffff
731 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
732 * @apptag_check_mask: check bitmask of application tag.
734 struct ib_t10_dif_domain {
735 enum ib_t10_dif_bg_type bg_type;
743 u16 apptag_check_mask;
747 * struct ib_sig_domain - Parameters for signature domain
748 * @sig_type: specific signauture type
749 * @sig: union of all signature domain attributes that may
750 * be used to set domain layout.
752 struct ib_sig_domain {
753 enum ib_signature_type sig_type;
755 struct ib_t10_dif_domain dif;
760 * struct ib_sig_attrs - Parameters for signature handover operation
761 * @check_mask: bitmask for signature byte check (8 bytes)
762 * @mem: memory domain layout desciptor.
763 * @wire: wire domain layout desciptor.
765 struct ib_sig_attrs {
767 struct ib_sig_domain mem;
768 struct ib_sig_domain wire;
771 enum ib_sig_err_type {
778 * struct ib_sig_err - signature error descriptor
781 enum ib_sig_err_type err_type;
788 enum ib_mr_status_check {
789 IB_MR_CHECK_SIG_STATUS = 1,
793 * struct ib_mr_status - Memory region status container
795 * @fail_status: Bitmask of MR checks status. For each
796 * failed check a corresponding status bit is set.
797 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
800 struct ib_mr_status {
802 struct ib_sig_err sig_err;
806 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
808 * @mult: multiple to convert.
810 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
813 struct ib_global_route grh;
827 IB_WC_LOC_EEC_OP_ERR,
832 IB_WC_LOC_ACCESS_ERR,
833 IB_WC_REM_INV_REQ_ERR,
834 IB_WC_REM_ACCESS_ERR,
837 IB_WC_RNR_RETRY_EXC_ERR,
838 IB_WC_LOC_RDD_VIOL_ERR,
839 IB_WC_REM_INV_RD_REQ_ERR,
842 IB_WC_INV_EEC_STATE_ERR,
844 IB_WC_RESP_TIMEOUT_ERR,
848 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
859 IB_WC_MASKED_COMP_SWAP,
860 IB_WC_MASKED_FETCH_ADD,
862 * Set value of IB_WC_RECV so consumers can test if a completion is a
863 * receive by testing (opcode & IB_WC_RECV).
866 IB_WC_RECV_RDMA_WITH_IMM
871 IB_WC_WITH_IMM = (1<<1),
872 IB_WC_WITH_INVALIDATE = (1<<2),
873 IB_WC_IP_CSUM_OK = (1<<3),
874 IB_WC_WITH_SMAC = (1<<4),
875 IB_WC_WITH_VLAN = (1<<5),
876 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
882 struct ib_cqe *wr_cqe;
884 enum ib_wc_status status;
885 enum ib_wc_opcode opcode;
899 u8 port_num; /* valid only for DR SMPs on switches */
905 enum ib_cq_notify_flags {
906 IB_CQ_SOLICITED = 1 << 0,
907 IB_CQ_NEXT_COMP = 1 << 1,
908 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
909 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
917 enum ib_srq_attr_mask {
918 IB_SRQ_MAX_WR = 1 << 0,
919 IB_SRQ_LIMIT = 1 << 1,
928 struct ib_srq_init_attr {
929 void (*event_handler)(struct ib_event *, void *);
931 struct ib_srq_attr attr;
932 enum ib_srq_type srq_type;
936 struct ib_xrcd *xrcd;
950 * Maximum number of rdma_rw_ctx structures in flight at a time.
951 * ib_create_qp() will calculate the right amount of neededed WRs
952 * and MRs based on this.
964 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
965 * here (and in that order) since the MAD layer uses them as
966 * indices into a 2-entry table.
975 IB_QPT_RAW_ETHERTYPE,
976 IB_QPT_RAW_PACKET = 8,
980 /* Reserve a range for qp types internal to the low level driver.
981 * These qp types will not be visible at the IB core layer, so the
982 * IB_QPT_MAX usages should not be affected in the core layer
984 IB_QPT_RESERVED1 = 0x1000,
996 enum ib_qp_create_flags {
997 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
998 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
999 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1000 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1001 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1002 IB_QP_CREATE_NETIF_QP = 1 << 5,
1003 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1004 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
1005 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1006 /* reserve bits 26-31 for low level drivers' internal use */
1007 IB_QP_CREATE_RESERVED_START = 1 << 26,
1008 IB_QP_CREATE_RESERVED_END = 1 << 31,
1012 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1013 * callback to destroy the passed in QP.
1016 struct ib_qp_init_attr {
1017 void (*event_handler)(struct ib_event *, void *);
1019 struct ib_cq *send_cq;
1020 struct ib_cq *recv_cq;
1022 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1023 struct ib_qp_cap cap;
1024 enum ib_sig_type sq_sig_type;
1025 enum ib_qp_type qp_type;
1026 enum ib_qp_create_flags create_flags;
1029 * Only needed for special QP types, or when using the RW API.
1032 struct ib_rwq_ind_table *rwq_ind_tbl;
1035 struct ib_qp_open_attr {
1036 void (*event_handler)(struct ib_event *, void *);
1039 enum ib_qp_type qp_type;
1042 enum ib_rnr_timeout {
1043 IB_RNR_TIMER_655_36 = 0,
1044 IB_RNR_TIMER_000_01 = 1,
1045 IB_RNR_TIMER_000_02 = 2,
1046 IB_RNR_TIMER_000_03 = 3,
1047 IB_RNR_TIMER_000_04 = 4,
1048 IB_RNR_TIMER_000_06 = 5,
1049 IB_RNR_TIMER_000_08 = 6,
1050 IB_RNR_TIMER_000_12 = 7,
1051 IB_RNR_TIMER_000_16 = 8,
1052 IB_RNR_TIMER_000_24 = 9,
1053 IB_RNR_TIMER_000_32 = 10,
1054 IB_RNR_TIMER_000_48 = 11,
1055 IB_RNR_TIMER_000_64 = 12,
1056 IB_RNR_TIMER_000_96 = 13,
1057 IB_RNR_TIMER_001_28 = 14,
1058 IB_RNR_TIMER_001_92 = 15,
1059 IB_RNR_TIMER_002_56 = 16,
1060 IB_RNR_TIMER_003_84 = 17,
1061 IB_RNR_TIMER_005_12 = 18,
1062 IB_RNR_TIMER_007_68 = 19,
1063 IB_RNR_TIMER_010_24 = 20,
1064 IB_RNR_TIMER_015_36 = 21,
1065 IB_RNR_TIMER_020_48 = 22,
1066 IB_RNR_TIMER_030_72 = 23,
1067 IB_RNR_TIMER_040_96 = 24,
1068 IB_RNR_TIMER_061_44 = 25,
1069 IB_RNR_TIMER_081_92 = 26,
1070 IB_RNR_TIMER_122_88 = 27,
1071 IB_RNR_TIMER_163_84 = 28,
1072 IB_RNR_TIMER_245_76 = 29,
1073 IB_RNR_TIMER_327_68 = 30,
1074 IB_RNR_TIMER_491_52 = 31
1077 enum ib_qp_attr_mask {
1079 IB_QP_CUR_STATE = (1<<1),
1080 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1081 IB_QP_ACCESS_FLAGS = (1<<3),
1082 IB_QP_PKEY_INDEX = (1<<4),
1083 IB_QP_PORT = (1<<5),
1084 IB_QP_QKEY = (1<<6),
1086 IB_QP_PATH_MTU = (1<<8),
1087 IB_QP_TIMEOUT = (1<<9),
1088 IB_QP_RETRY_CNT = (1<<10),
1089 IB_QP_RNR_RETRY = (1<<11),
1090 IB_QP_RQ_PSN = (1<<12),
1091 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1092 IB_QP_ALT_PATH = (1<<14),
1093 IB_QP_MIN_RNR_TIMER = (1<<15),
1094 IB_QP_SQ_PSN = (1<<16),
1095 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1096 IB_QP_PATH_MIG_STATE = (1<<18),
1097 IB_QP_CAP = (1<<19),
1098 IB_QP_DEST_QPN = (1<<20),
1099 IB_QP_RESERVED1 = (1<<21),
1100 IB_QP_RESERVED2 = (1<<22),
1101 IB_QP_RESERVED3 = (1<<23),
1102 IB_QP_RESERVED4 = (1<<24),
1127 enum ib_qp_state qp_state;
1128 enum ib_qp_state cur_qp_state;
1129 enum ib_mtu path_mtu;
1130 enum ib_mig_state path_mig_state;
1135 int qp_access_flags;
1136 struct ib_qp_cap cap;
1137 struct ib_ah_attr ah_attr;
1138 struct ib_ah_attr alt_ah_attr;
1141 u8 en_sqd_async_notify;
1144 u8 max_dest_rd_atomic;
1156 IB_WR_RDMA_WRITE_WITH_IMM,
1158 IB_WR_SEND_WITH_IMM,
1160 IB_WR_ATOMIC_CMP_AND_SWP,
1161 IB_WR_ATOMIC_FETCH_AND_ADD,
1163 IB_WR_SEND_WITH_INV,
1164 IB_WR_RDMA_READ_WITH_INV,
1167 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1168 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1170 /* reserve values for low level drivers' internal use.
1171 * These values will not be used at all in the ib core layer.
1173 IB_WR_RESERVED1 = 0xf0,
1185 enum ib_send_flags {
1187 IB_SEND_SIGNALED = (1<<1),
1188 IB_SEND_SOLICITED = (1<<2),
1189 IB_SEND_INLINE = (1<<3),
1190 IB_SEND_IP_CSUM = (1<<4),
1192 /* reserve bits 26-31 for low level drivers' internal use */
1193 IB_SEND_RESERVED_START = (1 << 26),
1194 IB_SEND_RESERVED_END = (1 << 31),
1204 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1208 struct ib_send_wr *next;
1211 struct ib_cqe *wr_cqe;
1213 struct ib_sge *sg_list;
1215 enum ib_wr_opcode opcode;
1219 u32 invalidate_rkey;
1224 struct ib_send_wr wr;
1229 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1231 return container_of(wr, struct ib_rdma_wr, wr);
1234 struct ib_atomic_wr {
1235 struct ib_send_wr wr;
1239 u64 compare_add_mask;
1244 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1246 return container_of(wr, struct ib_atomic_wr, wr);
1250 struct ib_send_wr wr;
1257 u16 pkey_index; /* valid for GSI only */
1258 u8 port_num; /* valid for DR SMPs on switch only */
1261 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1263 return container_of(wr, struct ib_ud_wr, wr);
1267 struct ib_send_wr wr;
1273 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1275 return container_of(wr, struct ib_reg_wr, wr);
1278 struct ib_sig_handover_wr {
1279 struct ib_send_wr wr;
1280 struct ib_sig_attrs *sig_attrs;
1281 struct ib_mr *sig_mr;
1283 struct ib_sge *prot;
1286 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1288 return container_of(wr, struct ib_sig_handover_wr, wr);
1292 struct ib_recv_wr *next;
1295 struct ib_cqe *wr_cqe;
1297 struct ib_sge *sg_list;
1301 enum ib_access_flags {
1302 IB_ACCESS_LOCAL_WRITE = 1,
1303 IB_ACCESS_REMOTE_WRITE = (1<<1),
1304 IB_ACCESS_REMOTE_READ = (1<<2),
1305 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1306 IB_ACCESS_MW_BIND = (1<<4),
1307 IB_ZERO_BASED = (1<<5),
1308 IB_ACCESS_ON_DEMAND = (1<<6),
1312 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1313 * are hidden here instead of a uapi header!
1315 enum ib_mr_rereg_flags {
1316 IB_MR_REREG_TRANS = 1,
1317 IB_MR_REREG_PD = (1<<1),
1318 IB_MR_REREG_ACCESS = (1<<2),
1319 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1322 struct ib_fmr_attr {
1330 struct ib_ucontext {
1331 struct ib_device *device;
1332 struct list_head pd_list;
1333 struct list_head mr_list;
1334 struct list_head mw_list;
1335 struct list_head cq_list;
1336 struct list_head qp_list;
1337 struct list_head srq_list;
1338 struct list_head ah_list;
1339 struct list_head xrcd_list;
1340 struct list_head rule_list;
1341 struct list_head wq_list;
1342 struct list_head rwq_ind_tbl_list;
1346 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1347 struct rb_root umem_tree;
1349 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1350 * mmu notifiers registration.
1352 struct rw_semaphore umem_rwsem;
1353 void (*invalidate_range)(struct ib_umem *umem,
1354 unsigned long start, unsigned long end);
1356 struct mmu_notifier mn;
1357 atomic_t notifier_count;
1358 /* A list of umems that don't have private mmu notifier counters yet. */
1359 struct list_head no_private_counters;
1365 u64 user_handle; /* handle given to us by userspace */
1366 struct ib_ucontext *context; /* associated user context */
1367 void *object; /* containing object */
1368 struct list_head list; /* link to context's list */
1369 int id; /* index into kernel idr */
1371 struct rw_semaphore mutex; /* protects .live */
1372 struct rcu_head rcu; /* kfree_rcu() overhead */
1377 const void __user *inbuf;
1378 void __user *outbuf;
1386 struct ib_device *device;
1387 struct ib_uobject *uobject;
1388 atomic_t usecnt; /* count all resources */
1390 u32 unsafe_global_rkey;
1393 * Implementation details of the RDMA core, don't use in drivers:
1395 struct ib_mr *__internal_mr;
1399 struct ib_device *device;
1400 atomic_t usecnt; /* count all exposed resources */
1401 struct inode *inode;
1403 struct mutex tgt_qp_mutex;
1404 struct list_head tgt_qp_list;
1408 struct ib_device *device;
1410 struct ib_uobject *uobject;
1413 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1415 enum ib_poll_context {
1416 IB_POLL_DIRECT, /* caller context, no hw completions */
1417 IB_POLL_SOFTIRQ, /* poll from softirq context */
1418 IB_POLL_WORKQUEUE, /* poll from workqueue */
1422 struct ib_device *device;
1423 struct ib_uobject *uobject;
1424 ib_comp_handler comp_handler;
1425 void (*event_handler)(struct ib_event *, void *);
1428 atomic_t usecnt; /* count number of work queues */
1429 enum ib_poll_context poll_ctx;
1432 struct irq_poll iop;
1433 struct work_struct work;
1438 struct ib_device *device;
1440 struct ib_uobject *uobject;
1441 void (*event_handler)(struct ib_event *, void *);
1443 enum ib_srq_type srq_type;
1448 struct ib_xrcd *xrcd;
1466 struct ib_device *device;
1467 struct ib_uobject *uobject;
1469 void (*event_handler)(struct ib_event *, void *);
1473 enum ib_wq_state state;
1474 enum ib_wq_type wq_type;
1478 struct ib_wq_init_attr {
1480 enum ib_wq_type wq_type;
1484 void (*event_handler)(struct ib_event *, void *);
1487 enum ib_wq_attr_mask {
1488 IB_WQ_STATE = 1 << 0,
1489 IB_WQ_CUR_STATE = 1 << 1,
1493 enum ib_wq_state wq_state;
1494 enum ib_wq_state curr_wq_state;
1497 struct ib_rwq_ind_table {
1498 struct ib_device *device;
1499 struct ib_uobject *uobject;
1502 u32 log_ind_tbl_size;
1503 struct ib_wq **ind_tbl;
1506 struct ib_rwq_ind_table_init_attr {
1507 u32 log_ind_tbl_size;
1508 /* Each entry is a pointer to Receive Work Queue */
1509 struct ib_wq **ind_tbl;
1513 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1514 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1517 struct ib_device *device;
1519 struct ib_cq *send_cq;
1520 struct ib_cq *recv_cq;
1523 struct list_head rdma_mrs;
1524 struct list_head sig_mrs;
1526 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1527 struct list_head xrcd_list;
1529 /* count times opened, mcast attaches, flow attaches */
1531 struct list_head open_list;
1532 struct ib_qp *real_qp;
1533 struct ib_uobject *uobject;
1534 void (*event_handler)(struct ib_event *, void *);
1539 enum ib_qp_type qp_type;
1540 struct ib_rwq_ind_table *rwq_ind_tbl;
1544 struct ib_device *device;
1550 unsigned int page_size;
1553 struct ib_uobject *uobject; /* user */
1554 struct list_head qp_entry; /* FR */
1559 struct ib_device *device;
1561 struct ib_uobject *uobject;
1563 enum ib_mw_type type;
1567 struct ib_device *device;
1569 struct list_head list;
1574 /* Supported steering options */
1575 enum ib_flow_attr_type {
1576 /* steering according to rule specifications */
1577 IB_FLOW_ATTR_NORMAL = 0x0,
1578 /* default unicast and multicast rule -
1579 * receive all Eth traffic which isn't steered to any QP
1581 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1582 /* default multicast rule -
1583 * receive all Eth multicast traffic which isn't steered to any QP
1585 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1586 /* sniffer rule - receive all port traffic */
1587 IB_FLOW_ATTR_SNIFFER = 0x3
1590 /* Supported steering header types */
1591 enum ib_flow_spec_type {
1593 IB_FLOW_SPEC_ETH = 0x20,
1594 IB_FLOW_SPEC_IB = 0x22,
1596 IB_FLOW_SPEC_IPV4 = 0x30,
1597 IB_FLOW_SPEC_IPV6 = 0x31,
1599 IB_FLOW_SPEC_TCP = 0x40,
1600 IB_FLOW_SPEC_UDP = 0x41
1602 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1603 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1605 /* Flow steering rule priority is set according to it's domain.
1606 * Lower domain value means higher priority.
1608 enum ib_flow_domain {
1609 IB_FLOW_DOMAIN_USER,
1610 IB_FLOW_DOMAIN_ETHTOOL,
1613 IB_FLOW_DOMAIN_NUM /* Must be last */
1616 enum ib_flow_flags {
1617 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1618 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1621 struct ib_flow_eth_filter {
1630 struct ib_flow_spec_eth {
1631 enum ib_flow_spec_type type;
1633 struct ib_flow_eth_filter val;
1634 struct ib_flow_eth_filter mask;
1637 struct ib_flow_ib_filter {
1644 struct ib_flow_spec_ib {
1645 enum ib_flow_spec_type type;
1647 struct ib_flow_ib_filter val;
1648 struct ib_flow_ib_filter mask;
1651 /* IPv4 header flags */
1652 enum ib_ipv4_flags {
1653 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1654 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1655 last have this flag set */
1658 struct ib_flow_ipv4_filter {
1669 struct ib_flow_spec_ipv4 {
1670 enum ib_flow_spec_type type;
1672 struct ib_flow_ipv4_filter val;
1673 struct ib_flow_ipv4_filter mask;
1676 struct ib_flow_ipv6_filter {
1687 struct ib_flow_spec_ipv6 {
1688 enum ib_flow_spec_type type;
1690 struct ib_flow_ipv6_filter val;
1691 struct ib_flow_ipv6_filter mask;
1694 struct ib_flow_tcp_udp_filter {
1701 struct ib_flow_spec_tcp_udp {
1702 enum ib_flow_spec_type type;
1704 struct ib_flow_tcp_udp_filter val;
1705 struct ib_flow_tcp_udp_filter mask;
1708 union ib_flow_spec {
1710 enum ib_flow_spec_type type;
1713 struct ib_flow_spec_eth eth;
1714 struct ib_flow_spec_ib ib;
1715 struct ib_flow_spec_ipv4 ipv4;
1716 struct ib_flow_spec_tcp_udp tcp_udp;
1717 struct ib_flow_spec_ipv6 ipv6;
1720 struct ib_flow_attr {
1721 enum ib_flow_attr_type type;
1727 /* Following are the optional layers according to user request
1728 * struct ib_flow_spec_xxx
1729 * struct ib_flow_spec_yyy
1735 struct ib_uobject *uobject;
1741 enum ib_process_mad_flags {
1742 IB_MAD_IGNORE_MKEY = 1,
1743 IB_MAD_IGNORE_BKEY = 2,
1744 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1747 enum ib_mad_result {
1748 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1749 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1750 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1751 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1754 #define IB_DEVICE_NAME_MAX 64
1758 struct ib_event_handler event_handler;
1759 struct ib_pkey_cache **pkey_cache;
1760 struct ib_gid_table **gid_cache;
1764 struct ib_dma_mapping_ops {
1765 int (*mapping_error)(struct ib_device *dev,
1767 u64 (*map_single)(struct ib_device *dev,
1768 void *ptr, size_t size,
1769 enum dma_data_direction direction);
1770 void (*unmap_single)(struct ib_device *dev,
1771 u64 addr, size_t size,
1772 enum dma_data_direction direction);
1773 u64 (*map_page)(struct ib_device *dev,
1774 struct page *page, unsigned long offset,
1776 enum dma_data_direction direction);
1777 void (*unmap_page)(struct ib_device *dev,
1778 u64 addr, size_t size,
1779 enum dma_data_direction direction);
1780 int (*map_sg)(struct ib_device *dev,
1781 struct scatterlist *sg, int nents,
1782 enum dma_data_direction direction);
1783 void (*unmap_sg)(struct ib_device *dev,
1784 struct scatterlist *sg, int nents,
1785 enum dma_data_direction direction);
1786 void (*sync_single_for_cpu)(struct ib_device *dev,
1789 enum dma_data_direction dir);
1790 void (*sync_single_for_device)(struct ib_device *dev,
1793 enum dma_data_direction dir);
1794 void *(*alloc_coherent)(struct ib_device *dev,
1798 void (*free_coherent)(struct ib_device *dev,
1799 size_t size, void *cpu_addr,
1805 struct ib_port_immutable {
1813 struct device *dma_device;
1815 char name[IB_DEVICE_NAME_MAX];
1817 struct list_head event_handler_list;
1818 spinlock_t event_handler_lock;
1820 spinlock_t client_data_lock;
1821 struct list_head core_list;
1822 /* Access to the client_data_list is protected by the client_data_lock
1823 * spinlock and the lists_rwsem read-write semaphore */
1824 struct list_head client_data_list;
1826 struct ib_cache cache;
1828 * port_immutable is indexed by port number
1830 struct ib_port_immutable *port_immutable;
1832 int num_comp_vectors;
1834 struct iw_cm_verbs *iwcm;
1837 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1838 * driver initialized data. The struct is kfree()'ed by the sysfs
1839 * core when the device is removed. A lifespan of -1 in the return
1840 * struct tells the core to set a default lifespan.
1842 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1845 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1846 * @index - The index in the value array we wish to have updated, or
1847 * num_counters if we want all stats updated
1849 * < 0 - Error, no counters updated
1850 * index - Updated the single counter pointed to by index
1851 * num_counters - Updated all counters (will reset the timestamp
1852 * and prevent further calls for lifespan milliseconds)
1853 * Drivers are allowed to update all counters in leiu of just the
1854 * one given in index at their option
1856 int (*get_hw_stats)(struct ib_device *device,
1857 struct rdma_hw_stats *stats,
1858 u8 port, int index);
1859 int (*query_device)(struct ib_device *device,
1860 struct ib_device_attr *device_attr,
1861 struct ib_udata *udata);
1862 int (*query_port)(struct ib_device *device,
1864 struct ib_port_attr *port_attr);
1865 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1867 /* When calling get_netdev, the HW vendor's driver should return the
1868 * net device of device @device at port @port_num or NULL if such
1869 * a net device doesn't exist. The vendor driver should call dev_hold
1870 * on this net device. The HW vendor's device driver must guarantee
1871 * that this function returns NULL before the net device reaches
1872 * NETDEV_UNREGISTER_FINAL state.
1874 struct net_device *(*get_netdev)(struct ib_device *device,
1876 int (*query_gid)(struct ib_device *device,
1877 u8 port_num, int index,
1879 /* When calling add_gid, the HW vendor's driver should
1880 * add the gid of device @device at gid index @index of
1881 * port @port_num to be @gid. Meta-info of that gid (for example,
1882 * the network device related to this gid is available
1883 * at @attr. @context allows the HW vendor driver to store extra
1884 * information together with a GID entry. The HW vendor may allocate
1885 * memory to contain this information and store it in @context when a
1886 * new GID entry is written to. Params are consistent until the next
1887 * call of add_gid or delete_gid. The function should return 0 on
1888 * success or error otherwise. The function could be called
1889 * concurrently for different ports. This function is only called
1890 * when roce_gid_table is used.
1892 int (*add_gid)(struct ib_device *device,
1895 const union ib_gid *gid,
1896 const struct ib_gid_attr *attr,
1898 /* When calling del_gid, the HW vendor's driver should delete the
1899 * gid of device @device at gid index @index of port @port_num.
1900 * Upon the deletion of a GID entry, the HW vendor must free any
1901 * allocated memory. The caller will clear @context afterwards.
1902 * This function is only called when roce_gid_table is used.
1904 int (*del_gid)(struct ib_device *device,
1908 int (*query_pkey)(struct ib_device *device,
1909 u8 port_num, u16 index, u16 *pkey);
1910 int (*modify_device)(struct ib_device *device,
1911 int device_modify_mask,
1912 struct ib_device_modify *device_modify);
1913 int (*modify_port)(struct ib_device *device,
1914 u8 port_num, int port_modify_mask,
1915 struct ib_port_modify *port_modify);
1916 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1917 struct ib_udata *udata);
1918 int (*dealloc_ucontext)(struct ib_ucontext *context);
1919 int (*mmap)(struct ib_ucontext *context,
1920 struct vm_area_struct *vma);
1921 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1922 struct ib_ucontext *context,
1923 struct ib_udata *udata);
1924 int (*dealloc_pd)(struct ib_pd *pd);
1925 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1926 struct ib_ah_attr *ah_attr);
1927 int (*modify_ah)(struct ib_ah *ah,
1928 struct ib_ah_attr *ah_attr);
1929 int (*query_ah)(struct ib_ah *ah,
1930 struct ib_ah_attr *ah_attr);
1931 int (*destroy_ah)(struct ib_ah *ah);
1932 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1933 struct ib_srq_init_attr *srq_init_attr,
1934 struct ib_udata *udata);
1935 int (*modify_srq)(struct ib_srq *srq,
1936 struct ib_srq_attr *srq_attr,
1937 enum ib_srq_attr_mask srq_attr_mask,
1938 struct ib_udata *udata);
1939 int (*query_srq)(struct ib_srq *srq,
1940 struct ib_srq_attr *srq_attr);
1941 int (*destroy_srq)(struct ib_srq *srq);
1942 int (*post_srq_recv)(struct ib_srq *srq,
1943 struct ib_recv_wr *recv_wr,
1944 struct ib_recv_wr **bad_recv_wr);
1945 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1946 struct ib_qp_init_attr *qp_init_attr,
1947 struct ib_udata *udata);
1948 int (*modify_qp)(struct ib_qp *qp,
1949 struct ib_qp_attr *qp_attr,
1951 struct ib_udata *udata);
1952 int (*query_qp)(struct ib_qp *qp,
1953 struct ib_qp_attr *qp_attr,
1955 struct ib_qp_init_attr *qp_init_attr);
1956 int (*destroy_qp)(struct ib_qp *qp);
1957 int (*post_send)(struct ib_qp *qp,
1958 struct ib_send_wr *send_wr,
1959 struct ib_send_wr **bad_send_wr);
1960 int (*post_recv)(struct ib_qp *qp,
1961 struct ib_recv_wr *recv_wr,
1962 struct ib_recv_wr **bad_recv_wr);
1963 struct ib_cq * (*create_cq)(struct ib_device *device,
1964 const struct ib_cq_init_attr *attr,
1965 struct ib_ucontext *context,
1966 struct ib_udata *udata);
1967 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1969 int (*destroy_cq)(struct ib_cq *cq);
1970 int (*resize_cq)(struct ib_cq *cq, int cqe,
1971 struct ib_udata *udata);
1972 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1974 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1975 int (*req_notify_cq)(struct ib_cq *cq,
1976 enum ib_cq_notify_flags flags);
1977 int (*req_ncomp_notif)(struct ib_cq *cq,
1979 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1980 int mr_access_flags);
1981 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1982 u64 start, u64 length,
1984 int mr_access_flags,
1985 struct ib_udata *udata);
1986 int (*rereg_user_mr)(struct ib_mr *mr,
1988 u64 start, u64 length,
1990 int mr_access_flags,
1992 struct ib_udata *udata);
1993 int (*dereg_mr)(struct ib_mr *mr);
1994 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1995 enum ib_mr_type mr_type,
1997 int (*map_mr_sg)(struct ib_mr *mr,
1998 struct scatterlist *sg,
2000 unsigned int *sg_offset);
2001 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2002 enum ib_mw_type type,
2003 struct ib_udata *udata);
2004 int (*dealloc_mw)(struct ib_mw *mw);
2005 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2006 int mr_access_flags,
2007 struct ib_fmr_attr *fmr_attr);
2008 int (*map_phys_fmr)(struct ib_fmr *fmr,
2009 u64 *page_list, int list_len,
2011 int (*unmap_fmr)(struct list_head *fmr_list);
2012 int (*dealloc_fmr)(struct ib_fmr *fmr);
2013 int (*attach_mcast)(struct ib_qp *qp,
2016 int (*detach_mcast)(struct ib_qp *qp,
2019 int (*process_mad)(struct ib_device *device,
2020 int process_mad_flags,
2022 const struct ib_wc *in_wc,
2023 const struct ib_grh *in_grh,
2024 const struct ib_mad_hdr *in_mad,
2026 struct ib_mad_hdr *out_mad,
2027 size_t *out_mad_size,
2028 u16 *out_mad_pkey_index);
2029 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2030 struct ib_ucontext *ucontext,
2031 struct ib_udata *udata);
2032 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2033 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2037 int (*destroy_flow)(struct ib_flow *flow_id);
2038 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2039 struct ib_mr_status *mr_status);
2040 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2041 void (*drain_rq)(struct ib_qp *qp);
2042 void (*drain_sq)(struct ib_qp *qp);
2043 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2045 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2046 struct ifla_vf_info *ivf);
2047 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2048 struct ifla_vf_stats *stats);
2049 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2051 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2052 struct ib_wq_init_attr *init_attr,
2053 struct ib_udata *udata);
2054 int (*destroy_wq)(struct ib_wq *wq);
2055 int (*modify_wq)(struct ib_wq *wq,
2056 struct ib_wq_attr *attr,
2058 struct ib_udata *udata);
2059 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2060 struct ib_rwq_ind_table_init_attr *init_attr,
2061 struct ib_udata *udata);
2062 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2063 struct ib_dma_mapping_ops *dma_ops;
2065 struct module *owner;
2067 struct kobject *ports_parent;
2068 struct list_head port_list;
2071 IB_DEV_UNINITIALIZED,
2077 u64 uverbs_cmd_mask;
2078 u64 uverbs_ex_cmd_mask;
2086 struct ib_device_attr attrs;
2087 struct attribute_group *hw_stats_ag;
2088 struct rdma_hw_stats *hw_stats;
2091 * The following mandatory functions are used only at device
2092 * registration. Keep functions such as these at the end of this
2093 * structure to avoid cache line misses when accessing struct ib_device
2096 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2097 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2102 void (*add) (struct ib_device *);
2103 void (*remove)(struct ib_device *, void *client_data);
2105 /* Returns the net_dev belonging to this ib_client and matching the
2107 * @dev: An RDMA device that the net_dev use for communication.
2108 * @port: A physical port number on the RDMA device.
2109 * @pkey: P_Key that the net_dev uses if applicable.
2110 * @gid: A GID that the net_dev uses to communicate.
2111 * @addr: An IP address the net_dev is configured with.
2112 * @client_data: The device's client data set by ib_set_client_data().
2114 * An ib_client that implements a net_dev on top of RDMA devices
2115 * (such as IP over IB) should implement this callback, allowing the
2116 * rdma_cm module to find the right net_dev for a given request.
2118 * The caller is responsible for calling dev_put on the returned
2120 struct net_device *(*get_net_dev_by_params)(
2121 struct ib_device *dev,
2124 const union ib_gid *gid,
2125 const struct sockaddr *addr,
2127 struct list_head list;
2130 struct ib_device *ib_alloc_device(size_t size);
2131 void ib_dealloc_device(struct ib_device *device);
2133 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2135 int ib_register_device(struct ib_device *device,
2136 int (*port_callback)(struct ib_device *,
2137 u8, struct kobject *));
2138 void ib_unregister_device(struct ib_device *device);
2140 int ib_register_client (struct ib_client *client);
2141 void ib_unregister_client(struct ib_client *client);
2143 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2144 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2147 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2149 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2152 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2154 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2157 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2161 const void __user *p = udata->inbuf + offset;
2165 if (len > USHRT_MAX)
2168 buf = kmalloc(len, GFP_KERNEL);
2172 if (copy_from_user(buf, p, len))
2175 ret = !memchr_inv(buf, 0, len);
2183 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2184 * contains all required attributes and no attributes not allowed for
2185 * the given QP state transition.
2186 * @cur_state: Current QP state
2187 * @next_state: Next QP state
2189 * @mask: Mask of supplied QP attributes
2190 * @ll : link layer of port
2192 * This function is a helper function that a low-level driver's
2193 * modify_qp method can use to validate the consumer's input. It
2194 * checks that cur_state and next_state are valid QP states, that a
2195 * transition from cur_state to next_state is allowed by the IB spec,
2196 * and that the attribute mask supplied is allowed for the transition.
2198 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2199 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2200 enum rdma_link_layer ll);
2202 int ib_register_event_handler (struct ib_event_handler *event_handler);
2203 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2204 void ib_dispatch_event(struct ib_event *event);
2206 int ib_query_port(struct ib_device *device,
2207 u8 port_num, struct ib_port_attr *port_attr);
2209 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2213 * rdma_cap_ib_switch - Check if the device is IB switch
2214 * @device: Device to check
2216 * Device driver is responsible for setting is_switch bit on
2217 * in ib_device structure at init time.
2219 * Return: true if the device is IB switch.
2221 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2223 return device->is_switch;
2227 * rdma_start_port - Return the first valid port number for the device
2230 * @device: Device to be checked
2232 * Return start port number
2234 static inline u8 rdma_start_port(const struct ib_device *device)
2236 return rdma_cap_ib_switch(device) ? 0 : 1;
2240 * rdma_end_port - Return the last valid port number for the device
2243 * @device: Device to be checked
2245 * Return last port number
2247 static inline u8 rdma_end_port(const struct ib_device *device)
2249 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2252 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2254 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2257 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2259 return device->port_immutable[port_num].core_cap_flags &
2260 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2263 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2265 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2268 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2270 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2273 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2275 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2278 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2280 return rdma_protocol_ib(device, port_num) ||
2281 rdma_protocol_roce(device, port_num);
2285 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2286 * Management Datagrams.
2287 * @device: Device to check
2288 * @port_num: Port number to check
2290 * Management Datagrams (MAD) are a required part of the InfiniBand
2291 * specification and are supported on all InfiniBand devices. A slightly
2292 * extended version are also supported on OPA interfaces.
2294 * Return: true if the port supports sending/receiving of MAD packets.
2296 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2298 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2302 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2303 * Management Datagrams.
2304 * @device: Device to check
2305 * @port_num: Port number to check
2307 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2308 * datagrams with their own versions. These OPA MADs share many but not all of
2309 * the characteristics of InfiniBand MADs.
2311 * OPA MADs differ in the following ways:
2313 * 1) MADs are variable size up to 2K
2314 * IBTA defined MADs remain fixed at 256 bytes
2315 * 2) OPA SMPs must carry valid PKeys
2316 * 3) OPA SMP packets are a different format
2318 * Return: true if the port supports OPA MAD packet formats.
2320 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2322 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2323 == RDMA_CORE_CAP_OPA_MAD;
2327 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2328 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2329 * @device: Device to check
2330 * @port_num: Port number to check
2332 * Each InfiniBand node is required to provide a Subnet Management Agent
2333 * that the subnet manager can access. Prior to the fabric being fully
2334 * configured by the subnet manager, the SMA is accessed via a well known
2335 * interface called the Subnet Management Interface (SMI). This interface
2336 * uses directed route packets to communicate with the SM to get around the
2337 * chicken and egg problem of the SM needing to know what's on the fabric
2338 * in order to configure the fabric, and needing to configure the fabric in
2339 * order to send packets to the devices on the fabric. These directed
2340 * route packets do not need the fabric fully configured in order to reach
2341 * their destination. The SMI is the only method allowed to send
2342 * directed route packets on an InfiniBand fabric.
2344 * Return: true if the port provides an SMI.
2346 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2348 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2352 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2353 * Communication Manager.
2354 * @device: Device to check
2355 * @port_num: Port number to check
2357 * The InfiniBand Communication Manager is one of many pre-defined General
2358 * Service Agents (GSA) that are accessed via the General Service
2359 * Interface (GSI). It's role is to facilitate establishment of connections
2360 * between nodes as well as other management related tasks for established
2363 * Return: true if the port supports an IB CM (this does not guarantee that
2364 * a CM is actually running however).
2366 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2368 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2372 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2373 * Communication Manager.
2374 * @device: Device to check
2375 * @port_num: Port number to check
2377 * Similar to above, but specific to iWARP connections which have a different
2378 * managment protocol than InfiniBand.
2380 * Return: true if the port supports an iWARP CM (this does not guarantee that
2381 * a CM is actually running however).
2383 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2385 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2389 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2390 * Subnet Administration.
2391 * @device: Device to check
2392 * @port_num: Port number to check
2394 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2395 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2396 * fabrics, devices should resolve routes to other hosts by contacting the
2397 * SA to query the proper route.
2399 * Return: true if the port should act as a client to the fabric Subnet
2400 * Administration interface. This does not imply that the SA service is
2403 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2405 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2409 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2411 * @device: Device to check
2412 * @port_num: Port number to check
2414 * InfiniBand multicast registration is more complex than normal IPv4 or
2415 * IPv6 multicast registration. Each Host Channel Adapter must register
2416 * with the Subnet Manager when it wishes to join a multicast group. It
2417 * should do so only once regardless of how many queue pairs it subscribes
2418 * to this group. And it should leave the group only after all queue pairs
2419 * attached to the group have been detached.
2421 * Return: true if the port must undertake the additional adminstrative
2422 * overhead of registering/unregistering with the SM and tracking of the
2423 * total number of queue pairs attached to the multicast group.
2425 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2427 return rdma_cap_ib_sa(device, port_num);
2431 * rdma_cap_af_ib - Check if the port of device has the capability
2432 * Native Infiniband Address.
2433 * @device: Device to check
2434 * @port_num: Port number to check
2436 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2437 * GID. RoCE uses a different mechanism, but still generates a GID via
2438 * a prescribed mechanism and port specific data.
2440 * Return: true if the port uses a GID address to identify devices on the
2443 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2445 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2449 * rdma_cap_eth_ah - Check if the port of device has the capability
2450 * Ethernet Address Handle.
2451 * @device: Device to check
2452 * @port_num: Port number to check
2454 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2455 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2456 * port. Normally, packet headers are generated by the sending host
2457 * adapter, but when sending connectionless datagrams, we must manually
2458 * inject the proper headers for the fabric we are communicating over.
2460 * Return: true if we are running as a RoCE port and must force the
2461 * addition of a Global Route Header built from our Ethernet Address
2462 * Handle into our header list for connectionless packets.
2464 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2466 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2470 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2473 * @port_num: Port number
2475 * This MAD size includes the MAD headers and MAD payload. No other headers
2478 * Return the max MAD size required by the Port. Will return 0 if the port
2479 * does not support MADs
2481 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2483 return device->port_immutable[port_num].max_mad_size;
2487 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2488 * @device: Device to check
2489 * @port_num: Port number to check
2491 * RoCE GID table mechanism manages the various GIDs for a device.
2493 * NOTE: if allocating the port's GID table has failed, this call will still
2494 * return true, but any RoCE GID table API will fail.
2496 * Return: true if the port uses RoCE GID table mechanism in order to manage
2499 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2502 return rdma_protocol_roce(device, port_num) &&
2503 device->add_gid && device->del_gid;
2507 * Check if the device supports READ W/ INVALIDATE.
2509 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2512 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2513 * has support for it yet.
2515 return rdma_protocol_iwarp(dev, port_num);
2518 int ib_query_gid(struct ib_device *device,
2519 u8 port_num, int index, union ib_gid *gid,
2520 struct ib_gid_attr *attr);
2522 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2524 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2525 struct ifla_vf_info *info);
2526 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2527 struct ifla_vf_stats *stats);
2528 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2531 int ib_query_pkey(struct ib_device *device,
2532 u8 port_num, u16 index, u16 *pkey);
2534 int ib_modify_device(struct ib_device *device,
2535 int device_modify_mask,
2536 struct ib_device_modify *device_modify);
2538 int ib_modify_port(struct ib_device *device,
2539 u8 port_num, int port_modify_mask,
2540 struct ib_port_modify *port_modify);
2542 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2543 enum ib_gid_type gid_type, struct net_device *ndev,
2544 u8 *port_num, u16 *index);
2546 int ib_find_pkey(struct ib_device *device,
2547 u8 port_num, u16 pkey, u16 *index);
2551 * Create a memory registration for all memory in the system and place
2552 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2553 * ULPs to avoid the overhead of dynamic MRs.
2555 * This flag is generally considered unsafe and must only be used in
2556 * extremly trusted environments. Every use of it will log a warning
2557 * in the kernel log.
2559 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2562 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2563 const char *caller);
2564 #define ib_alloc_pd(device, flags) \
2565 __ib_alloc_pd((device), (flags), __func__)
2566 void ib_dealloc_pd(struct ib_pd *pd);
2569 * ib_create_ah - Creates an address handle for the given address vector.
2570 * @pd: The protection domain associated with the address handle.
2571 * @ah_attr: The attributes of the address vector.
2573 * The address handle is used to reference a local or global destination
2574 * in all UD QP post sends.
2576 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2579 * ib_init_ah_from_wc - Initializes address handle attributes from a
2581 * @device: Device on which the received message arrived.
2582 * @port_num: Port on which the received message arrived.
2583 * @wc: Work completion associated with the received message.
2584 * @grh: References the received global route header. This parameter is
2585 * ignored unless the work completion indicates that the GRH is valid.
2586 * @ah_attr: Returned attributes that can be used when creating an address
2587 * handle for replying to the message.
2589 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2590 const struct ib_wc *wc, const struct ib_grh *grh,
2591 struct ib_ah_attr *ah_attr);
2594 * ib_create_ah_from_wc - Creates an address handle associated with the
2595 * sender of the specified work completion.
2596 * @pd: The protection domain associated with the address handle.
2597 * @wc: Work completion information associated with a received message.
2598 * @grh: References the received global route header. This parameter is
2599 * ignored unless the work completion indicates that the GRH is valid.
2600 * @port_num: The outbound port number to associate with the address.
2602 * The address handle is used to reference a local or global destination
2603 * in all UD QP post sends.
2605 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2606 const struct ib_grh *grh, u8 port_num);
2609 * ib_modify_ah - Modifies the address vector associated with an address
2611 * @ah: The address handle to modify.
2612 * @ah_attr: The new address vector attributes to associate with the
2615 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2618 * ib_query_ah - Queries the address vector associated with an address
2620 * @ah: The address handle to query.
2621 * @ah_attr: The address vector attributes associated with the address
2624 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2627 * ib_destroy_ah - Destroys an address handle.
2628 * @ah: The address handle to destroy.
2630 int ib_destroy_ah(struct ib_ah *ah);
2633 * ib_create_srq - Creates a SRQ associated with the specified protection
2635 * @pd: The protection domain associated with the SRQ.
2636 * @srq_init_attr: A list of initial attributes required to create the
2637 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2638 * the actual capabilities of the created SRQ.
2640 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2641 * requested size of the SRQ, and set to the actual values allocated
2642 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2643 * will always be at least as large as the requested values.
2645 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2646 struct ib_srq_init_attr *srq_init_attr);
2649 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2650 * @srq: The SRQ to modify.
2651 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2652 * the current values of selected SRQ attributes are returned.
2653 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2654 * are being modified.
2656 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2657 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2658 * the number of receives queued drops below the limit.
2660 int ib_modify_srq(struct ib_srq *srq,
2661 struct ib_srq_attr *srq_attr,
2662 enum ib_srq_attr_mask srq_attr_mask);
2665 * ib_query_srq - Returns the attribute list and current values for the
2667 * @srq: The SRQ to query.
2668 * @srq_attr: The attributes of the specified SRQ.
2670 int ib_query_srq(struct ib_srq *srq,
2671 struct ib_srq_attr *srq_attr);
2674 * ib_destroy_srq - Destroys the specified SRQ.
2675 * @srq: The SRQ to destroy.
2677 int ib_destroy_srq(struct ib_srq *srq);
2680 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2681 * @srq: The SRQ to post the work request on.
2682 * @recv_wr: A list of work requests to post on the receive queue.
2683 * @bad_recv_wr: On an immediate failure, this parameter will reference
2684 * the work request that failed to be posted on the QP.
2686 static inline int ib_post_srq_recv(struct ib_srq *srq,
2687 struct ib_recv_wr *recv_wr,
2688 struct ib_recv_wr **bad_recv_wr)
2690 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2694 * ib_create_qp - Creates a QP associated with the specified protection
2696 * @pd: The protection domain associated with the QP.
2697 * @qp_init_attr: A list of initial attributes required to create the
2698 * QP. If QP creation succeeds, then the attributes are updated to
2699 * the actual capabilities of the created QP.
2701 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2702 struct ib_qp_init_attr *qp_init_attr);
2705 * ib_modify_qp - Modifies the attributes for the specified QP and then
2706 * transitions the QP to the given state.
2707 * @qp: The QP to modify.
2708 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2709 * the current values of selected QP attributes are returned.
2710 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2711 * are being modified.
2713 int ib_modify_qp(struct ib_qp *qp,
2714 struct ib_qp_attr *qp_attr,
2718 * ib_query_qp - Returns the attribute list and current values for the
2720 * @qp: The QP to query.
2721 * @qp_attr: The attributes of the specified QP.
2722 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2723 * @qp_init_attr: Additional attributes of the selected QP.
2725 * The qp_attr_mask may be used to limit the query to gathering only the
2726 * selected attributes.
2728 int ib_query_qp(struct ib_qp *qp,
2729 struct ib_qp_attr *qp_attr,
2731 struct ib_qp_init_attr *qp_init_attr);
2734 * ib_destroy_qp - Destroys the specified QP.
2735 * @qp: The QP to destroy.
2737 int ib_destroy_qp(struct ib_qp *qp);
2740 * ib_open_qp - Obtain a reference to an existing sharable QP.
2741 * @xrcd - XRC domain
2742 * @qp_open_attr: Attributes identifying the QP to open.
2744 * Returns a reference to a sharable QP.
2746 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2747 struct ib_qp_open_attr *qp_open_attr);
2750 * ib_close_qp - Release an external reference to a QP.
2751 * @qp: The QP handle to release
2753 * The opened QP handle is released by the caller. The underlying
2754 * shared QP is not destroyed until all internal references are released.
2756 int ib_close_qp(struct ib_qp *qp);
2759 * ib_post_send - Posts a list of work requests to the send queue of
2761 * @qp: The QP to post the work request on.
2762 * @send_wr: A list of work requests to post on the send queue.
2763 * @bad_send_wr: On an immediate failure, this parameter will reference
2764 * the work request that failed to be posted on the QP.
2766 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2767 * error is returned, the QP state shall not be affected,
2768 * ib_post_send() will return an immediate error after queueing any
2769 * earlier work requests in the list.
2771 static inline int ib_post_send(struct ib_qp *qp,
2772 struct ib_send_wr *send_wr,
2773 struct ib_send_wr **bad_send_wr)
2775 return qp->device->post_send(qp, send_wr, bad_send_wr);
2779 * ib_post_recv - Posts a list of work requests to the receive queue of
2781 * @qp: The QP to post the work request on.
2782 * @recv_wr: A list of work requests to post on the receive queue.
2783 * @bad_recv_wr: On an immediate failure, this parameter will reference
2784 * the work request that failed to be posted on the QP.
2786 static inline int ib_post_recv(struct ib_qp *qp,
2787 struct ib_recv_wr *recv_wr,
2788 struct ib_recv_wr **bad_recv_wr)
2790 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2793 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2794 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2795 void ib_free_cq(struct ib_cq *cq);
2796 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2799 * ib_create_cq - Creates a CQ on the specified device.
2800 * @device: The device on which to create the CQ.
2801 * @comp_handler: A user-specified callback that is invoked when a
2802 * completion event occurs on the CQ.
2803 * @event_handler: A user-specified callback that is invoked when an
2804 * asynchronous event not associated with a completion occurs on the CQ.
2805 * @cq_context: Context associated with the CQ returned to the user via
2806 * the associated completion and event handlers.
2807 * @cq_attr: The attributes the CQ should be created upon.
2809 * Users can examine the cq structure to determine the actual CQ size.
2811 struct ib_cq *ib_create_cq(struct ib_device *device,
2812 ib_comp_handler comp_handler,
2813 void (*event_handler)(struct ib_event *, void *),
2815 const struct ib_cq_init_attr *cq_attr);
2818 * ib_resize_cq - Modifies the capacity of the CQ.
2819 * @cq: The CQ to resize.
2820 * @cqe: The minimum size of the CQ.
2822 * Users can examine the cq structure to determine the actual CQ size.
2824 int ib_resize_cq(struct ib_cq *cq, int cqe);
2827 * ib_modify_cq - Modifies moderation params of the CQ
2828 * @cq: The CQ to modify.
2829 * @cq_count: number of CQEs that will trigger an event
2830 * @cq_period: max period of time in usec before triggering an event
2833 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2836 * ib_destroy_cq - Destroys the specified CQ.
2837 * @cq: The CQ to destroy.
2839 int ib_destroy_cq(struct ib_cq *cq);
2842 * ib_poll_cq - poll a CQ for completion(s)
2843 * @cq:the CQ being polled
2844 * @num_entries:maximum number of completions to return
2845 * @wc:array of at least @num_entries &struct ib_wc where completions
2848 * Poll a CQ for (possibly multiple) completions. If the return value
2849 * is < 0, an error occurred. If the return value is >= 0, it is the
2850 * number of completions returned. If the return value is
2851 * non-negative and < num_entries, then the CQ was emptied.
2853 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2856 return cq->device->poll_cq(cq, num_entries, wc);
2860 * ib_peek_cq - Returns the number of unreaped completions currently
2861 * on the specified CQ.
2862 * @cq: The CQ to peek.
2863 * @wc_cnt: A minimum number of unreaped completions to check for.
2865 * If the number of unreaped completions is greater than or equal to wc_cnt,
2866 * this function returns wc_cnt, otherwise, it returns the actual number of
2867 * unreaped completions.
2869 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2872 * ib_req_notify_cq - Request completion notification on a CQ.
2873 * @cq: The CQ to generate an event for.
2875 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2876 * to request an event on the next solicited event or next work
2877 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2878 * may also be |ed in to request a hint about missed events, as
2882 * < 0 means an error occurred while requesting notification
2883 * == 0 means notification was requested successfully, and if
2884 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2885 * were missed and it is safe to wait for another event. In
2886 * this case is it guaranteed that any work completions added
2887 * to the CQ since the last CQ poll will trigger a completion
2888 * notification event.
2889 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2890 * in. It means that the consumer must poll the CQ again to
2891 * make sure it is empty to avoid missing an event because of a
2892 * race between requesting notification and an entry being
2893 * added to the CQ. This return value means it is possible
2894 * (but not guaranteed) that a work completion has been added
2895 * to the CQ since the last poll without triggering a
2896 * completion notification event.
2898 static inline int ib_req_notify_cq(struct ib_cq *cq,
2899 enum ib_cq_notify_flags flags)
2901 return cq->device->req_notify_cq(cq, flags);
2905 * ib_req_ncomp_notif - Request completion notification when there are
2906 * at least the specified number of unreaped completions on the CQ.
2907 * @cq: The CQ to generate an event for.
2908 * @wc_cnt: The number of unreaped completions that should be on the
2909 * CQ before an event is generated.
2911 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2913 return cq->device->req_ncomp_notif ?
2914 cq->device->req_ncomp_notif(cq, wc_cnt) :
2919 * ib_dma_mapping_error - check a DMA addr for error
2920 * @dev: The device for which the dma_addr was created
2921 * @dma_addr: The DMA address to check
2923 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2926 return dev->dma_ops->mapping_error(dev, dma_addr);
2927 return dma_mapping_error(dev->dma_device, dma_addr);
2931 * ib_dma_map_single - Map a kernel virtual address to DMA address
2932 * @dev: The device for which the dma_addr is to be created
2933 * @cpu_addr: The kernel virtual address
2934 * @size: The size of the region in bytes
2935 * @direction: The direction of the DMA
2937 static inline u64 ib_dma_map_single(struct ib_device *dev,
2938 void *cpu_addr, size_t size,
2939 enum dma_data_direction direction)
2942 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2943 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2947 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2948 * @dev: The device for which the DMA address was created
2949 * @addr: The DMA address
2950 * @size: The size of the region in bytes
2951 * @direction: The direction of the DMA
2953 static inline void ib_dma_unmap_single(struct ib_device *dev,
2954 u64 addr, size_t size,
2955 enum dma_data_direction direction)
2958 dev->dma_ops->unmap_single(dev, addr, size, direction);
2960 dma_unmap_single(dev->dma_device, addr, size, direction);
2963 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2964 void *cpu_addr, size_t size,
2965 enum dma_data_direction direction,
2966 unsigned long dma_attrs)
2968 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2969 direction, dma_attrs);
2972 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2973 u64 addr, size_t size,
2974 enum dma_data_direction direction,
2975 unsigned long dma_attrs)
2977 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2978 direction, dma_attrs);
2982 * ib_dma_map_page - Map a physical page to DMA address
2983 * @dev: The device for which the dma_addr is to be created
2984 * @page: The page to be mapped
2985 * @offset: The offset within the page
2986 * @size: The size of the region in bytes
2987 * @direction: The direction of the DMA
2989 static inline u64 ib_dma_map_page(struct ib_device *dev,
2991 unsigned long offset,
2993 enum dma_data_direction direction)
2996 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2997 return dma_map_page(dev->dma_device, page, offset, size, direction);
3001 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3002 * @dev: The device for which the DMA address was created
3003 * @addr: The DMA address
3004 * @size: The size of the region in bytes
3005 * @direction: The direction of the DMA
3007 static inline void ib_dma_unmap_page(struct ib_device *dev,
3008 u64 addr, size_t size,
3009 enum dma_data_direction direction)
3012 dev->dma_ops->unmap_page(dev, addr, size, direction);
3014 dma_unmap_page(dev->dma_device, addr, size, direction);
3018 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3019 * @dev: The device for which the DMA addresses are to be created
3020 * @sg: The array of scatter/gather entries
3021 * @nents: The number of scatter/gather entries
3022 * @direction: The direction of the DMA
3024 static inline int ib_dma_map_sg(struct ib_device *dev,
3025 struct scatterlist *sg, int nents,
3026 enum dma_data_direction direction)
3029 return dev->dma_ops->map_sg(dev, sg, nents, direction);
3030 return dma_map_sg(dev->dma_device, sg, nents, direction);
3034 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3035 * @dev: The device for which the DMA addresses were created
3036 * @sg: The array of scatter/gather entries
3037 * @nents: The number of scatter/gather entries
3038 * @direction: The direction of the DMA
3040 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3041 struct scatterlist *sg, int nents,
3042 enum dma_data_direction direction)
3045 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3047 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3050 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3051 struct scatterlist *sg, int nents,
3052 enum dma_data_direction direction,
3053 unsigned long dma_attrs)
3055 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3059 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3060 struct scatterlist *sg, int nents,
3061 enum dma_data_direction direction,
3062 unsigned long dma_attrs)
3064 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3067 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3068 * @dev: The device for which the DMA addresses were created
3069 * @sg: The scatter/gather entry
3071 * Note: this function is obsolete. To do: change all occurrences of
3072 * ib_sg_dma_address() into sg_dma_address().
3074 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3075 struct scatterlist *sg)
3077 return sg_dma_address(sg);
3081 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3082 * @dev: The device for which the DMA addresses were created
3083 * @sg: The scatter/gather entry
3085 * Note: this function is obsolete. To do: change all occurrences of
3086 * ib_sg_dma_len() into sg_dma_len().
3088 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3089 struct scatterlist *sg)
3091 return sg_dma_len(sg);
3095 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3096 * @dev: The device for which the DMA address was created
3097 * @addr: The DMA address
3098 * @size: The size of the region in bytes
3099 * @dir: The direction of the DMA
3101 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3104 enum dma_data_direction dir)
3107 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3109 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3113 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3114 * @dev: The device for which the DMA address was created
3115 * @addr: The DMA address
3116 * @size: The size of the region in bytes
3117 * @dir: The direction of the DMA
3119 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3122 enum dma_data_direction dir)
3125 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3127 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3131 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3132 * @dev: The device for which the DMA address is requested
3133 * @size: The size of the region to allocate in bytes
3134 * @dma_handle: A pointer for returning the DMA address of the region
3135 * @flag: memory allocator flags
3137 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3143 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3148 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3149 *dma_handle = handle;
3155 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3156 * @dev: The device for which the DMA addresses were allocated
3157 * @size: The size of the region
3158 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3159 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3161 static inline void ib_dma_free_coherent(struct ib_device *dev,
3162 size_t size, void *cpu_addr,
3166 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3168 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3172 * ib_dereg_mr - Deregisters a memory region and removes it from the
3173 * HCA translation table.
3174 * @mr: The memory region to deregister.
3176 * This function can fail, if the memory region has memory windows bound to it.
3178 int ib_dereg_mr(struct ib_mr *mr);
3180 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3181 enum ib_mr_type mr_type,
3185 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3187 * @mr - struct ib_mr pointer to be updated.
3188 * @newkey - new key to be used.
3190 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3192 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3193 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3197 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3198 * for calculating a new rkey for type 2 memory windows.
3199 * @rkey - the rkey to increment.
3201 static inline u32 ib_inc_rkey(u32 rkey)
3203 const u32 mask = 0x000000ff;
3204 return ((rkey + 1) & mask) | (rkey & ~mask);
3208 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3209 * @pd: The protection domain associated with the unmapped region.
3210 * @mr_access_flags: Specifies the memory access rights.
3211 * @fmr_attr: Attributes of the unmapped region.
3213 * A fast memory region must be mapped before it can be used as part of
3216 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3217 int mr_access_flags,
3218 struct ib_fmr_attr *fmr_attr);
3221 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3222 * @fmr: The fast memory region to associate with the pages.
3223 * @page_list: An array of physical pages to map to the fast memory region.
3224 * @list_len: The number of pages in page_list.
3225 * @iova: The I/O virtual address to use with the mapped region.
3227 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3228 u64 *page_list, int list_len,
3231 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3235 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3236 * @fmr_list: A linked list of fast memory regions to unmap.
3238 int ib_unmap_fmr(struct list_head *fmr_list);
3241 * ib_dealloc_fmr - Deallocates a fast memory region.
3242 * @fmr: The fast memory region to deallocate.
3244 int ib_dealloc_fmr(struct ib_fmr *fmr);
3247 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3248 * @qp: QP to attach to the multicast group. The QP must be type
3250 * @gid: Multicast group GID.
3251 * @lid: Multicast group LID in host byte order.
3253 * In order to send and receive multicast packets, subnet
3254 * administration must have created the multicast group and configured
3255 * the fabric appropriately. The port associated with the specified
3256 * QP must also be a member of the multicast group.
3258 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3261 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3262 * @qp: QP to detach from the multicast group.
3263 * @gid: Multicast group GID.
3264 * @lid: Multicast group LID in host byte order.
3266 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3269 * ib_alloc_xrcd - Allocates an XRC domain.
3270 * @device: The device on which to allocate the XRC domain.
3272 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3275 * ib_dealloc_xrcd - Deallocates an XRC domain.
3276 * @xrcd: The XRC domain to deallocate.
3278 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3280 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3281 struct ib_flow_attr *flow_attr, int domain);
3282 int ib_destroy_flow(struct ib_flow *flow_id);
3284 static inline int ib_check_mr_access(int flags)
3287 * Local write permission is required if remote write or
3288 * remote atomic permission is also requested.
3290 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3291 !(flags & IB_ACCESS_LOCAL_WRITE))
3298 * ib_check_mr_status: lightweight check of MR status.
3299 * This routine may provide status checks on a selected
3300 * ib_mr. first use is for signature status check.
3302 * @mr: A memory region.
3303 * @check_mask: Bitmask of which checks to perform from
3304 * ib_mr_status_check enumeration.
3305 * @mr_status: The container of relevant status checks.
3306 * failed checks will be indicated in the status bitmask
3307 * and the relevant info shall be in the error item.
3309 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3310 struct ib_mr_status *mr_status);
3312 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3313 u16 pkey, const union ib_gid *gid,
3314 const struct sockaddr *addr);
3315 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3316 struct ib_wq_init_attr *init_attr);
3317 int ib_destroy_wq(struct ib_wq *wq);
3318 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3320 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3321 struct ib_rwq_ind_table_init_attr*
3322 wq_ind_table_init_attr);
3323 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3325 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3326 unsigned int *sg_offset, unsigned int page_size);
3329 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3330 unsigned int *sg_offset, unsigned int page_size)
3334 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3340 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3341 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3343 void ib_drain_rq(struct ib_qp *qp);
3344 void ib_drain_sq(struct ib_qp *qp);
3345 void ib_drain_qp(struct ib_qp *qp);
3346 #endif /* IB_VERBS_H */