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>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
67 #include <uapi/rdma/rdma_user_ioctl.h>
68 #include <uapi/rdma/ib_user_ioctl_verbs.h>
70 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
72 extern struct workqueue_struct *ib_wq;
73 extern struct workqueue_struct *ib_comp_wq;
83 extern union ib_gid zgid;
86 /* If link layer is Ethernet, this is RoCE V1 */
89 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
93 #define ROCE_V2_UDP_DPORT 4791
95 struct net_device *ndev;
96 struct ib_device *device;
98 enum ib_gid_type gid_type;
103 enum rdma_node_type {
104 /* IB values map to NodeInfo:NodeType. */
114 /* set the local administered indication */
115 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
118 enum rdma_transport_type {
120 RDMA_TRANSPORT_IWARP,
121 RDMA_TRANSPORT_USNIC,
122 RDMA_TRANSPORT_USNIC_UDP
125 enum rdma_protocol_type {
129 RDMA_PROTOCOL_USNIC_UDP
132 __attribute_const__ enum rdma_transport_type
133 rdma_node_get_transport(enum rdma_node_type node_type);
135 enum rdma_network_type {
137 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
144 if (network_type == RDMA_NETWORK_IPV4 ||
145 network_type == RDMA_NETWORK_IPV6)
146 return IB_GID_TYPE_ROCE_UDP_ENCAP;
148 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
149 return IB_GID_TYPE_IB;
152 static inline enum rdma_network_type
153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
155 if (attr->gid_type == IB_GID_TYPE_IB)
156 return RDMA_NETWORK_IB;
158 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
159 return RDMA_NETWORK_IPV4;
161 return RDMA_NETWORK_IPV6;
164 enum rdma_link_layer {
165 IB_LINK_LAYER_UNSPECIFIED,
166 IB_LINK_LAYER_INFINIBAND,
167 IB_LINK_LAYER_ETHERNET,
170 enum ib_device_cap_flags {
171 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
172 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
173 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
174 IB_DEVICE_RAW_MULTI = (1 << 3),
175 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
176 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
177 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
178 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
179 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
180 /* Not in use, former INIT_TYPE = (1 << 9),*/
181 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
182 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
183 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
184 IB_DEVICE_SRQ_RESIZE = (1 << 13),
185 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
188 * This device supports a per-device lkey or stag that can be
189 * used without performing a memory registration for the local
190 * memory. Note that ULPs should never check this flag, but
191 * instead of use the local_dma_lkey flag in the ib_pd structure,
192 * which will always contain a usable lkey.
194 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
195 /* Reserved, old SEND_W_INV = (1 << 16),*/
196 IB_DEVICE_MEM_WINDOW = (1 << 17),
198 * Devices should set IB_DEVICE_UD_IP_SUM if they support
199 * insertion of UDP and TCP checksum on outgoing UD IPoIB
200 * messages and can verify the validity of checksum for
201 * incoming messages. Setting this flag implies that the
202 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
204 IB_DEVICE_UD_IP_CSUM = (1 << 18),
205 IB_DEVICE_UD_TSO = (1 << 19),
206 IB_DEVICE_XRC = (1 << 20),
209 * This device supports the IB "base memory management extension",
210 * which includes support for fast registrations (IB_WR_REG_MR,
211 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
212 * also be set by any iWarp device which must support FRs to comply
213 * to the iWarp verbs spec. iWarp devices also support the
214 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
217 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
218 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
219 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
220 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
221 IB_DEVICE_RC_IP_CSUM = (1 << 25),
222 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
223 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
225 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
226 * support execution of WQEs that involve synchronization
227 * of I/O operations with single completion queue managed
230 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
231 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
232 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
233 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
234 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
235 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
236 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
237 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
238 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
239 /* The device supports padding incoming writes to cacheline. */
240 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
243 enum ib_signature_prot_cap {
244 IB_PROT_T10DIF_TYPE_1 = 1,
245 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
246 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
249 enum ib_signature_guard_cap {
250 IB_GUARD_T10DIF_CRC = 1,
251 IB_GUARD_T10DIF_CSUM = 1 << 1,
260 enum ib_odp_general_cap_bits {
261 IB_ODP_SUPPORT = 1 << 0,
262 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
265 enum ib_odp_transport_cap_bits {
266 IB_ODP_SUPPORT_SEND = 1 << 0,
267 IB_ODP_SUPPORT_RECV = 1 << 1,
268 IB_ODP_SUPPORT_WRITE = 1 << 2,
269 IB_ODP_SUPPORT_READ = 1 << 3,
270 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
274 uint64_t general_caps;
276 uint32_t rc_odp_caps;
277 uint32_t uc_odp_caps;
278 uint32_t ud_odp_caps;
279 } per_transport_caps;
283 /* Corresponding bit will be set if qp type from
284 * 'enum ib_qp_type' is supported, e.g.
285 * supported_qpts |= 1 << IB_QPT_UD
288 u32 max_rwq_indirection_tables;
289 u32 max_rwq_indirection_table_size;
292 enum ib_tm_cap_flags {
293 /* Support tag matching on RC transport */
294 IB_TM_CAP_RC = 1 << 0,
298 /* Max size of RNDV header */
299 u32 max_rndv_hdr_size;
300 /* Max number of entries in tag matching list */
302 /* From enum ib_tm_cap_flags */
304 /* Max number of outstanding list operations */
306 /* Max number of SGE in tag matching entry */
310 struct ib_cq_init_attr {
316 enum ib_cq_attr_mask {
317 IB_CQ_MODERATE = 1 << 0,
321 u16 max_cq_moderation_count;
322 u16 max_cq_moderation_period;
325 struct ib_dm_mr_attr {
331 struct ib_dm_alloc_attr {
337 struct ib_device_attr {
339 __be64 sys_image_guid;
347 u64 device_cap_flags;
358 int max_qp_init_rd_atom;
359 int max_ee_init_rd_atom;
360 enum ib_atomic_cap atomic_cap;
361 enum ib_atomic_cap masked_atomic_cap;
368 int max_mcast_qp_attach;
369 int max_total_mcast_qp_attach;
376 unsigned int max_fast_reg_page_list_len;
378 u8 local_ca_ack_delay;
381 struct ib_odp_caps odp_caps;
382 uint64_t timestamp_mask;
383 uint64_t hca_core_clock; /* in KHZ */
384 struct ib_rss_caps rss_caps;
386 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
387 struct ib_tm_caps tm_caps;
388 struct ib_cq_caps cq_caps;
400 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
403 case IB_MTU_256: return 256;
404 case IB_MTU_512: return 512;
405 case IB_MTU_1024: return 1024;
406 case IB_MTU_2048: return 2048;
407 case IB_MTU_4096: return 4096;
412 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
416 else if (mtu >= 2048)
418 else if (mtu >= 1024)
432 IB_PORT_ACTIVE_DEFER = 5
442 static inline int ib_width_enum_to_int(enum ib_port_width width)
445 case IB_WIDTH_1X: return 1;
446 case IB_WIDTH_4X: return 4;
447 case IB_WIDTH_8X: return 8;
448 case IB_WIDTH_12X: return 12;
464 * struct rdma_hw_stats
465 * @lock - Mutex to protect parallel write access to lifespan and values
466 * of counters, which are 64bits and not guaranteeed to be written
467 * atomicaly on 32bits systems.
468 * @timestamp - Used by the core code to track when the last update was
469 * @lifespan - Used by the core code to determine how old the counters
470 * should be before being updated again. Stored in jiffies, defaults
471 * to 10 milliseconds, drivers can override the default be specifying
472 * their own value during their allocation routine.
473 * @name - Array of pointers to static names used for the counters in
475 * @num_counters - How many hardware counters there are. If name is
476 * shorter than this number, a kernel oops will result. Driver authors
477 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
478 * in their code to prevent this.
479 * @value - Array of u64 counters that are accessed by the sysfs code and
480 * filled in by the drivers get_stats routine
482 struct rdma_hw_stats {
483 struct mutex lock; /* Protect lifespan and values[] */
484 unsigned long timestamp;
485 unsigned long lifespan;
486 const char * const *names;
491 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
493 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
495 * @names - Array of static const char *
496 * @num_counters - How many elements in array
497 * @lifespan - How many milliseconds between updates
499 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
500 const char * const *names, int num_counters,
501 unsigned long lifespan)
503 struct rdma_hw_stats *stats;
505 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
509 stats->names = names;
510 stats->num_counters = num_counters;
511 stats->lifespan = msecs_to_jiffies(lifespan);
517 /* Define bits for the various functionality this port needs to be supported by
520 /* Management 0x00000FFF */
521 #define RDMA_CORE_CAP_IB_MAD 0x00000001
522 #define RDMA_CORE_CAP_IB_SMI 0x00000002
523 #define RDMA_CORE_CAP_IB_CM 0x00000004
524 #define RDMA_CORE_CAP_IW_CM 0x00000008
525 #define RDMA_CORE_CAP_IB_SA 0x00000010
526 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
528 /* Address format 0x000FF000 */
529 #define RDMA_CORE_CAP_AF_IB 0x00001000
530 #define RDMA_CORE_CAP_ETH_AH 0x00002000
531 #define RDMA_CORE_CAP_OPA_AH 0x00004000
532 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
534 /* Protocol 0xFFF00000 */
535 #define RDMA_CORE_CAP_PROT_IB 0x00100000
536 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
537 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
538 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
539 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
540 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
542 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
543 | RDMA_CORE_CAP_PROT_ROCE \
544 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
546 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
547 | RDMA_CORE_CAP_IB_MAD \
548 | RDMA_CORE_CAP_IB_SMI \
549 | RDMA_CORE_CAP_IB_CM \
550 | RDMA_CORE_CAP_IB_SA \
551 | RDMA_CORE_CAP_AF_IB)
552 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
553 | RDMA_CORE_CAP_IB_MAD \
554 | RDMA_CORE_CAP_IB_CM \
555 | RDMA_CORE_CAP_AF_IB \
556 | RDMA_CORE_CAP_ETH_AH)
557 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
558 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
559 | RDMA_CORE_CAP_IB_MAD \
560 | RDMA_CORE_CAP_IB_CM \
561 | RDMA_CORE_CAP_AF_IB \
562 | RDMA_CORE_CAP_ETH_AH)
563 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
564 | RDMA_CORE_CAP_IW_CM)
565 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
566 | RDMA_CORE_CAP_OPA_MAD)
568 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
570 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
572 struct ib_port_attr {
574 enum ib_port_state state;
576 enum ib_mtu active_mtu;
578 unsigned int ip_gids:1;
579 /* This is the value from PortInfo CapabilityMask, defined by IBA */
597 enum ib_device_modify_flags {
598 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
599 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
602 #define IB_DEVICE_NODE_DESC_MAX 64
604 struct ib_device_modify {
606 char node_desc[IB_DEVICE_NODE_DESC_MAX];
609 enum ib_port_modify_flags {
610 IB_PORT_SHUTDOWN = 1,
611 IB_PORT_INIT_TYPE = (1<<2),
612 IB_PORT_RESET_QKEY_CNTR = (1<<3),
613 IB_PORT_OPA_MASK_CHG = (1<<4)
616 struct ib_port_modify {
617 u32 set_port_cap_mask;
618 u32 clr_port_cap_mask;
626 IB_EVENT_QP_ACCESS_ERR,
630 IB_EVENT_PATH_MIG_ERR,
631 IB_EVENT_DEVICE_FATAL,
632 IB_EVENT_PORT_ACTIVE,
635 IB_EVENT_PKEY_CHANGE,
638 IB_EVENT_SRQ_LIMIT_REACHED,
639 IB_EVENT_QP_LAST_WQE_REACHED,
640 IB_EVENT_CLIENT_REREGISTER,
645 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
648 struct ib_device *device;
656 enum ib_event_type event;
659 struct ib_event_handler {
660 struct ib_device *device;
661 void (*handler)(struct ib_event_handler *, struct ib_event *);
662 struct list_head list;
665 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
667 (_ptr)->device = _device; \
668 (_ptr)->handler = _handler; \
669 INIT_LIST_HEAD(&(_ptr)->list); \
672 struct ib_global_route {
673 const struct ib_gid_attr *sgid_attr;
682 __be32 version_tclass_flow;
690 union rdma_network_hdr {
693 /* The IB spec states that if it's IPv4, the header
694 * is located in the last 20 bytes of the header.
697 struct iphdr roce4grh;
701 #define IB_QPN_MASK 0xFFFFFF
704 IB_MULTICAST_QPN = 0xffffff
707 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
708 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
715 IB_RATE_PORT_CURRENT = 0,
716 IB_RATE_2_5_GBPS = 2,
724 IB_RATE_120_GBPS = 10,
725 IB_RATE_14_GBPS = 11,
726 IB_RATE_56_GBPS = 12,
727 IB_RATE_112_GBPS = 13,
728 IB_RATE_168_GBPS = 14,
729 IB_RATE_25_GBPS = 15,
730 IB_RATE_100_GBPS = 16,
731 IB_RATE_200_GBPS = 17,
732 IB_RATE_300_GBPS = 18
736 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
737 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
738 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
739 * @rate: rate to convert.
741 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
744 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
745 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
746 * @rate: rate to convert.
748 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
752 * enum ib_mr_type - memory region type
753 * @IB_MR_TYPE_MEM_REG: memory region that is used for
754 * normal registration
755 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
756 * signature operations (data-integrity
758 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
759 * register any arbitrary sg lists (without
760 * the normal mr constraints - see
765 IB_MR_TYPE_SIGNATURE,
771 * IB_SIG_TYPE_NONE: Unprotected.
772 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
774 enum ib_signature_type {
780 * Signature T10-DIF block-guard types
781 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
782 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
784 enum ib_t10_dif_bg_type {
790 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
792 * @bg_type: T10-DIF block guard type (CRC|CSUM)
793 * @pi_interval: protection information interval.
794 * @bg: seed of guard computation.
795 * @app_tag: application tag of guard block
796 * @ref_tag: initial guard block reference tag.
797 * @ref_remap: Indicate wethear the reftag increments each block
798 * @app_escape: Indicate to skip block check if apptag=0xffff
799 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
800 * @apptag_check_mask: check bitmask of application tag.
802 struct ib_t10_dif_domain {
803 enum ib_t10_dif_bg_type bg_type;
811 u16 apptag_check_mask;
815 * struct ib_sig_domain - Parameters for signature domain
816 * @sig_type: specific signauture type
817 * @sig: union of all signature domain attributes that may
818 * be used to set domain layout.
820 struct ib_sig_domain {
821 enum ib_signature_type sig_type;
823 struct ib_t10_dif_domain dif;
828 * struct ib_sig_attrs - Parameters for signature handover operation
829 * @check_mask: bitmask for signature byte check (8 bytes)
830 * @mem: memory domain layout desciptor.
831 * @wire: wire domain layout desciptor.
833 struct ib_sig_attrs {
835 struct ib_sig_domain mem;
836 struct ib_sig_domain wire;
839 enum ib_sig_err_type {
846 * Signature check masks (8 bytes in total) according to the T10-PI standard:
847 * -------- -------- ------------
848 * | GUARD | APPTAG | REFTAG |
850 * -------- -------- ------------
853 IB_SIG_CHECK_GUARD = 0xc0,
854 IB_SIG_CHECK_APPTAG = 0x30,
855 IB_SIG_CHECK_REFTAG = 0x0f,
859 * struct ib_sig_err - signature error descriptor
862 enum ib_sig_err_type err_type;
869 enum ib_mr_status_check {
870 IB_MR_CHECK_SIG_STATUS = 1,
874 * struct ib_mr_status - Memory region status container
876 * @fail_status: Bitmask of MR checks status. For each
877 * failed check a corresponding status bit is set.
878 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
881 struct ib_mr_status {
883 struct ib_sig_err sig_err;
887 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
889 * @mult: multiple to convert.
891 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
893 enum rdma_ah_attr_type {
894 RDMA_AH_ATTR_TYPE_UNDEFINED,
895 RDMA_AH_ATTR_TYPE_IB,
896 RDMA_AH_ATTR_TYPE_ROCE,
897 RDMA_AH_ATTR_TYPE_OPA,
905 struct roce_ah_attr {
915 struct rdma_ah_attr {
916 struct ib_global_route grh;
921 enum rdma_ah_attr_type type;
923 struct ib_ah_attr ib;
924 struct roce_ah_attr roce;
925 struct opa_ah_attr opa;
933 IB_WC_LOC_EEC_OP_ERR,
938 IB_WC_LOC_ACCESS_ERR,
939 IB_WC_REM_INV_REQ_ERR,
940 IB_WC_REM_ACCESS_ERR,
943 IB_WC_RNR_RETRY_EXC_ERR,
944 IB_WC_LOC_RDD_VIOL_ERR,
945 IB_WC_REM_INV_RD_REQ_ERR,
948 IB_WC_INV_EEC_STATE_ERR,
950 IB_WC_RESP_TIMEOUT_ERR,
954 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
965 IB_WC_MASKED_COMP_SWAP,
966 IB_WC_MASKED_FETCH_ADD,
968 * Set value of IB_WC_RECV so consumers can test if a completion is a
969 * receive by testing (opcode & IB_WC_RECV).
972 IB_WC_RECV_RDMA_WITH_IMM
977 IB_WC_WITH_IMM = (1<<1),
978 IB_WC_WITH_INVALIDATE = (1<<2),
979 IB_WC_IP_CSUM_OK = (1<<3),
980 IB_WC_WITH_SMAC = (1<<4),
981 IB_WC_WITH_VLAN = (1<<5),
982 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
988 struct ib_cqe *wr_cqe;
990 enum ib_wc_status status;
991 enum ib_wc_opcode opcode;
1005 u8 port_num; /* valid only for DR SMPs on switches */
1008 u8 network_hdr_type;
1011 enum ib_cq_notify_flags {
1012 IB_CQ_SOLICITED = 1 << 0,
1013 IB_CQ_NEXT_COMP = 1 << 1,
1014 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1015 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1024 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1026 return srq_type == IB_SRQT_XRC ||
1027 srq_type == IB_SRQT_TM;
1030 enum ib_srq_attr_mask {
1031 IB_SRQ_MAX_WR = 1 << 0,
1032 IB_SRQ_LIMIT = 1 << 1,
1035 struct ib_srq_attr {
1041 struct ib_srq_init_attr {
1042 void (*event_handler)(struct ib_event *, void *);
1044 struct ib_srq_attr attr;
1045 enum ib_srq_type srq_type;
1051 struct ib_xrcd *xrcd;
1066 u32 max_inline_data;
1069 * Maximum number of rdma_rw_ctx structures in flight at a time.
1070 * ib_create_qp() will calculate the right amount of neededed WRs
1071 * and MRs based on this.
1083 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1084 * here (and in that order) since the MAD layer uses them as
1085 * indices into a 2-entry table.
1094 IB_QPT_RAW_ETHERTYPE,
1095 IB_QPT_RAW_PACKET = 8,
1099 IB_QPT_DRIVER = 0xFF,
1100 /* Reserve a range for qp types internal to the low level driver.
1101 * These qp types will not be visible at the IB core layer, so the
1102 * IB_QPT_MAX usages should not be affected in the core layer
1104 IB_QPT_RESERVED1 = 0x1000,
1116 enum ib_qp_create_flags {
1117 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1118 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1119 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1120 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1121 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1122 IB_QP_CREATE_NETIF_QP = 1 << 5,
1123 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1124 /* FREE = 1 << 7, */
1125 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1126 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1127 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1128 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1129 /* reserve bits 26-31 for low level drivers' internal use */
1130 IB_QP_CREATE_RESERVED_START = 1 << 26,
1131 IB_QP_CREATE_RESERVED_END = 1 << 31,
1135 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1136 * callback to destroy the passed in QP.
1139 struct ib_qp_init_attr {
1140 void (*event_handler)(struct ib_event *, void *);
1142 struct ib_cq *send_cq;
1143 struct ib_cq *recv_cq;
1145 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1146 struct ib_qp_cap cap;
1147 enum ib_sig_type sq_sig_type;
1148 enum ib_qp_type qp_type;
1149 enum ib_qp_create_flags create_flags;
1152 * Only needed for special QP types, or when using the RW API.
1155 struct ib_rwq_ind_table *rwq_ind_tbl;
1159 struct ib_qp_open_attr {
1160 void (*event_handler)(struct ib_event *, void *);
1163 enum ib_qp_type qp_type;
1166 enum ib_rnr_timeout {
1167 IB_RNR_TIMER_655_36 = 0,
1168 IB_RNR_TIMER_000_01 = 1,
1169 IB_RNR_TIMER_000_02 = 2,
1170 IB_RNR_TIMER_000_03 = 3,
1171 IB_RNR_TIMER_000_04 = 4,
1172 IB_RNR_TIMER_000_06 = 5,
1173 IB_RNR_TIMER_000_08 = 6,
1174 IB_RNR_TIMER_000_12 = 7,
1175 IB_RNR_TIMER_000_16 = 8,
1176 IB_RNR_TIMER_000_24 = 9,
1177 IB_RNR_TIMER_000_32 = 10,
1178 IB_RNR_TIMER_000_48 = 11,
1179 IB_RNR_TIMER_000_64 = 12,
1180 IB_RNR_TIMER_000_96 = 13,
1181 IB_RNR_TIMER_001_28 = 14,
1182 IB_RNR_TIMER_001_92 = 15,
1183 IB_RNR_TIMER_002_56 = 16,
1184 IB_RNR_TIMER_003_84 = 17,
1185 IB_RNR_TIMER_005_12 = 18,
1186 IB_RNR_TIMER_007_68 = 19,
1187 IB_RNR_TIMER_010_24 = 20,
1188 IB_RNR_TIMER_015_36 = 21,
1189 IB_RNR_TIMER_020_48 = 22,
1190 IB_RNR_TIMER_030_72 = 23,
1191 IB_RNR_TIMER_040_96 = 24,
1192 IB_RNR_TIMER_061_44 = 25,
1193 IB_RNR_TIMER_081_92 = 26,
1194 IB_RNR_TIMER_122_88 = 27,
1195 IB_RNR_TIMER_163_84 = 28,
1196 IB_RNR_TIMER_245_76 = 29,
1197 IB_RNR_TIMER_327_68 = 30,
1198 IB_RNR_TIMER_491_52 = 31
1201 enum ib_qp_attr_mask {
1203 IB_QP_CUR_STATE = (1<<1),
1204 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1205 IB_QP_ACCESS_FLAGS = (1<<3),
1206 IB_QP_PKEY_INDEX = (1<<4),
1207 IB_QP_PORT = (1<<5),
1208 IB_QP_QKEY = (1<<6),
1210 IB_QP_PATH_MTU = (1<<8),
1211 IB_QP_TIMEOUT = (1<<9),
1212 IB_QP_RETRY_CNT = (1<<10),
1213 IB_QP_RNR_RETRY = (1<<11),
1214 IB_QP_RQ_PSN = (1<<12),
1215 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1216 IB_QP_ALT_PATH = (1<<14),
1217 IB_QP_MIN_RNR_TIMER = (1<<15),
1218 IB_QP_SQ_PSN = (1<<16),
1219 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1220 IB_QP_PATH_MIG_STATE = (1<<18),
1221 IB_QP_CAP = (1<<19),
1222 IB_QP_DEST_QPN = (1<<20),
1223 IB_QP_RESERVED1 = (1<<21),
1224 IB_QP_RESERVED2 = (1<<22),
1225 IB_QP_RESERVED3 = (1<<23),
1226 IB_QP_RESERVED4 = (1<<24),
1227 IB_QP_RATE_LIMIT = (1<<25),
1252 enum ib_qp_state qp_state;
1253 enum ib_qp_state cur_qp_state;
1254 enum ib_mtu path_mtu;
1255 enum ib_mig_state path_mig_state;
1260 int qp_access_flags;
1261 struct ib_qp_cap cap;
1262 struct rdma_ah_attr ah_attr;
1263 struct rdma_ah_attr alt_ah_attr;
1266 u8 en_sqd_async_notify;
1269 u8 max_dest_rd_atomic;
1282 IB_WR_RDMA_WRITE_WITH_IMM,
1284 IB_WR_SEND_WITH_IMM,
1286 IB_WR_ATOMIC_CMP_AND_SWP,
1287 IB_WR_ATOMIC_FETCH_AND_ADD,
1289 IB_WR_SEND_WITH_INV,
1290 IB_WR_RDMA_READ_WITH_INV,
1293 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1294 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1296 /* reserve values for low level drivers' internal use.
1297 * These values will not be used at all in the ib core layer.
1299 IB_WR_RESERVED1 = 0xf0,
1311 enum ib_send_flags {
1313 IB_SEND_SIGNALED = (1<<1),
1314 IB_SEND_SOLICITED = (1<<2),
1315 IB_SEND_INLINE = (1<<3),
1316 IB_SEND_IP_CSUM = (1<<4),
1318 /* reserve bits 26-31 for low level drivers' internal use */
1319 IB_SEND_RESERVED_START = (1 << 26),
1320 IB_SEND_RESERVED_END = (1 << 31),
1330 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1334 struct ib_send_wr *next;
1337 struct ib_cqe *wr_cqe;
1339 struct ib_sge *sg_list;
1341 enum ib_wr_opcode opcode;
1345 u32 invalidate_rkey;
1350 struct ib_send_wr wr;
1355 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1357 return container_of(wr, struct ib_rdma_wr, wr);
1360 struct ib_atomic_wr {
1361 struct ib_send_wr wr;
1365 u64 compare_add_mask;
1370 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1372 return container_of(wr, struct ib_atomic_wr, wr);
1376 struct ib_send_wr wr;
1383 u16 pkey_index; /* valid for GSI only */
1384 u8 port_num; /* valid for DR SMPs on switch only */
1387 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1389 return container_of(wr, struct ib_ud_wr, wr);
1393 struct ib_send_wr wr;
1399 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1401 return container_of(wr, struct ib_reg_wr, wr);
1404 struct ib_sig_handover_wr {
1405 struct ib_send_wr wr;
1406 struct ib_sig_attrs *sig_attrs;
1407 struct ib_mr *sig_mr;
1409 struct ib_sge *prot;
1412 static inline const struct ib_sig_handover_wr *
1413 sig_handover_wr(const struct ib_send_wr *wr)
1415 return container_of(wr, struct ib_sig_handover_wr, wr);
1419 struct ib_recv_wr *next;
1422 struct ib_cqe *wr_cqe;
1424 struct ib_sge *sg_list;
1428 enum ib_access_flags {
1429 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1430 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1431 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1432 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1433 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1434 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1435 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1436 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1438 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1442 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1443 * are hidden here instead of a uapi header!
1445 enum ib_mr_rereg_flags {
1446 IB_MR_REREG_TRANS = 1,
1447 IB_MR_REREG_PD = (1<<1),
1448 IB_MR_REREG_ACCESS = (1<<2),
1449 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1452 struct ib_fmr_attr {
1460 enum rdma_remove_reason {
1462 * Userspace requested uobject deletion or initial try
1463 * to remove uobject via cleanup. Call could fail
1465 RDMA_REMOVE_DESTROY,
1466 /* Context deletion. This call should delete the actual object itself */
1468 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1469 RDMA_REMOVE_DRIVER_REMOVE,
1470 /* uobj is being cleaned-up before being committed */
1474 struct ib_rdmacg_object {
1475 #ifdef CONFIG_CGROUP_RDMA
1476 struct rdma_cgroup *cg; /* owner rdma cgroup */
1480 struct ib_ucontext {
1481 struct ib_device *device;
1482 struct ib_uverbs_file *ufile;
1484 * 'closing' can be read by the driver only during a destroy callback,
1485 * it is set when we are closing the file descriptor and indicates
1486 * that mm_sem may be locked.
1490 bool cleanup_retryable;
1493 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1494 struct rb_root_cached umem_tree;
1496 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1497 * mmu notifiers registration.
1499 struct rw_semaphore umem_rwsem;
1500 void (*invalidate_range)(struct ib_umem *umem,
1501 unsigned long start, unsigned long end);
1503 struct mmu_notifier mn;
1504 atomic_t notifier_count;
1505 /* A list of umems that don't have private mmu notifier counters yet. */
1506 struct list_head no_private_counters;
1510 struct ib_rdmacg_object cg_obj;
1514 u64 user_handle; /* handle given to us by userspace */
1515 /* ufile & ucontext owning this object */
1516 struct ib_uverbs_file *ufile;
1517 /* FIXME, save memory: ufile->context == context */
1518 struct ib_ucontext *context; /* associated user context */
1519 void *object; /* containing object */
1520 struct list_head list; /* link to context's list */
1521 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1522 int id; /* index into kernel idr */
1524 atomic_t usecnt; /* protects exclusive access */
1525 struct rcu_head rcu; /* kfree_rcu() overhead */
1527 const struct uverbs_api_object *uapi_object;
1531 const void __user *inbuf;
1532 void __user *outbuf;
1540 struct ib_device *device;
1541 struct ib_uobject *uobject;
1542 atomic_t usecnt; /* count all resources */
1544 u32 unsafe_global_rkey;
1547 * Implementation details of the RDMA core, don't use in drivers:
1549 struct ib_mr *__internal_mr;
1550 struct rdma_restrack_entry res;
1554 struct ib_device *device;
1555 atomic_t usecnt; /* count all exposed resources */
1556 struct inode *inode;
1558 struct mutex tgt_qp_mutex;
1559 struct list_head tgt_qp_list;
1563 struct ib_device *device;
1565 struct ib_uobject *uobject;
1566 const struct ib_gid_attr *sgid_attr;
1567 enum rdma_ah_attr_type type;
1570 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1572 enum ib_poll_context {
1573 IB_POLL_DIRECT, /* caller context, no hw completions */
1574 IB_POLL_SOFTIRQ, /* poll from softirq context */
1575 IB_POLL_WORKQUEUE, /* poll from workqueue */
1579 struct ib_device *device;
1580 struct ib_uobject *uobject;
1581 ib_comp_handler comp_handler;
1582 void (*event_handler)(struct ib_event *, void *);
1585 atomic_t usecnt; /* count number of work queues */
1586 enum ib_poll_context poll_ctx;
1589 struct irq_poll iop;
1590 struct work_struct work;
1593 * Implementation details of the RDMA core, don't use in drivers:
1595 struct rdma_restrack_entry res;
1599 struct ib_device *device;
1601 struct ib_uobject *uobject;
1602 void (*event_handler)(struct ib_event *, void *);
1604 enum ib_srq_type srq_type;
1611 struct ib_xrcd *xrcd;
1618 enum ib_raw_packet_caps {
1619 /* Strip cvlan from incoming packet and report it in the matching work
1620 * completion is supported.
1622 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1623 /* Scatter FCS field of an incoming packet to host memory is supported.
1625 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1626 /* Checksum offloads are supported (for both send and receive). */
1627 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1628 /* When a packet is received for an RQ with no receive WQEs, the
1629 * packet processing is delayed.
1631 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1645 struct ib_device *device;
1646 struct ib_uobject *uobject;
1648 void (*event_handler)(struct ib_event *, void *);
1652 enum ib_wq_state state;
1653 enum ib_wq_type wq_type;
1658 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1659 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1660 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1661 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1664 struct ib_wq_init_attr {
1666 enum ib_wq_type wq_type;
1670 void (*event_handler)(struct ib_event *, void *);
1671 u32 create_flags; /* Use enum ib_wq_flags */
1674 enum ib_wq_attr_mask {
1675 IB_WQ_STATE = 1 << 0,
1676 IB_WQ_CUR_STATE = 1 << 1,
1677 IB_WQ_FLAGS = 1 << 2,
1681 enum ib_wq_state wq_state;
1682 enum ib_wq_state curr_wq_state;
1683 u32 flags; /* Use enum ib_wq_flags */
1684 u32 flags_mask; /* Use enum ib_wq_flags */
1687 struct ib_rwq_ind_table {
1688 struct ib_device *device;
1689 struct ib_uobject *uobject;
1692 u32 log_ind_tbl_size;
1693 struct ib_wq **ind_tbl;
1696 struct ib_rwq_ind_table_init_attr {
1697 u32 log_ind_tbl_size;
1698 /* Each entry is a pointer to Receive Work Queue */
1699 struct ib_wq **ind_tbl;
1702 enum port_pkey_state {
1703 IB_PORT_PKEY_NOT_VALID = 0,
1704 IB_PORT_PKEY_VALID = 1,
1705 IB_PORT_PKEY_LISTED = 2,
1708 struct ib_qp_security;
1710 struct ib_port_pkey {
1711 enum port_pkey_state state;
1714 struct list_head qp_list;
1715 struct list_head to_error_list;
1716 struct ib_qp_security *sec;
1719 struct ib_ports_pkeys {
1720 struct ib_port_pkey main;
1721 struct ib_port_pkey alt;
1724 struct ib_qp_security {
1726 struct ib_device *dev;
1727 /* Hold this mutex when changing port and pkey settings. */
1729 struct ib_ports_pkeys *ports_pkeys;
1730 /* A list of all open shared QP handles. Required to enforce security
1731 * properly for all users of a shared QP.
1733 struct list_head shared_qp_list;
1736 atomic_t error_list_count;
1737 struct completion error_complete;
1738 int error_comps_pending;
1742 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1743 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1746 struct ib_device *device;
1748 struct ib_cq *send_cq;
1749 struct ib_cq *recv_cq;
1752 struct list_head rdma_mrs;
1753 struct list_head sig_mrs;
1755 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1756 struct list_head xrcd_list;
1758 /* count times opened, mcast attaches, flow attaches */
1760 struct list_head open_list;
1761 struct ib_qp *real_qp;
1762 struct ib_uobject *uobject;
1763 void (*event_handler)(struct ib_event *, void *);
1765 /* sgid_attrs associated with the AV's */
1766 const struct ib_gid_attr *av_sgid_attr;
1767 const struct ib_gid_attr *alt_path_sgid_attr;
1771 enum ib_qp_type qp_type;
1772 struct ib_rwq_ind_table *rwq_ind_tbl;
1773 struct ib_qp_security *qp_sec;
1777 * Implementation details of the RDMA core, don't use in drivers:
1779 struct rdma_restrack_entry res;
1783 struct ib_device *device;
1786 struct ib_uobject *uobject;
1791 struct ib_device *device;
1797 unsigned int page_size;
1800 struct ib_uobject *uobject; /* user */
1801 struct list_head qp_entry; /* FR */
1807 * Implementation details of the RDMA core, don't use in drivers:
1809 struct rdma_restrack_entry res;
1813 struct ib_device *device;
1815 struct ib_uobject *uobject;
1817 enum ib_mw_type type;
1821 struct ib_device *device;
1823 struct list_head list;
1828 /* Supported steering options */
1829 enum ib_flow_attr_type {
1830 /* steering according to rule specifications */
1831 IB_FLOW_ATTR_NORMAL = 0x0,
1832 /* default unicast and multicast rule -
1833 * receive all Eth traffic which isn't steered to any QP
1835 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1836 /* default multicast rule -
1837 * receive all Eth multicast traffic which isn't steered to any QP
1839 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1840 /* sniffer rule - receive all port traffic */
1841 IB_FLOW_ATTR_SNIFFER = 0x3
1844 /* Supported steering header types */
1845 enum ib_flow_spec_type {
1847 IB_FLOW_SPEC_ETH = 0x20,
1848 IB_FLOW_SPEC_IB = 0x22,
1850 IB_FLOW_SPEC_IPV4 = 0x30,
1851 IB_FLOW_SPEC_IPV6 = 0x31,
1852 IB_FLOW_SPEC_ESP = 0x34,
1854 IB_FLOW_SPEC_TCP = 0x40,
1855 IB_FLOW_SPEC_UDP = 0x41,
1856 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1857 IB_FLOW_SPEC_GRE = 0x51,
1858 IB_FLOW_SPEC_MPLS = 0x60,
1859 IB_FLOW_SPEC_INNER = 0x100,
1861 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1862 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1863 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1864 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1866 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1867 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1869 /* Flow steering rule priority is set according to it's domain.
1870 * Lower domain value means higher priority.
1872 enum ib_flow_domain {
1873 IB_FLOW_DOMAIN_USER,
1874 IB_FLOW_DOMAIN_ETHTOOL,
1877 IB_FLOW_DOMAIN_NUM /* Must be last */
1880 enum ib_flow_flags {
1881 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1882 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1883 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1886 struct ib_flow_eth_filter {
1895 struct ib_flow_spec_eth {
1898 struct ib_flow_eth_filter val;
1899 struct ib_flow_eth_filter mask;
1902 struct ib_flow_ib_filter {
1909 struct ib_flow_spec_ib {
1912 struct ib_flow_ib_filter val;
1913 struct ib_flow_ib_filter mask;
1916 /* IPv4 header flags */
1917 enum ib_ipv4_flags {
1918 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1919 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1920 last have this flag set */
1923 struct ib_flow_ipv4_filter {
1934 struct ib_flow_spec_ipv4 {
1937 struct ib_flow_ipv4_filter val;
1938 struct ib_flow_ipv4_filter mask;
1941 struct ib_flow_ipv6_filter {
1952 struct ib_flow_spec_ipv6 {
1955 struct ib_flow_ipv6_filter val;
1956 struct ib_flow_ipv6_filter mask;
1959 struct ib_flow_tcp_udp_filter {
1966 struct ib_flow_spec_tcp_udp {
1969 struct ib_flow_tcp_udp_filter val;
1970 struct ib_flow_tcp_udp_filter mask;
1973 struct ib_flow_tunnel_filter {
1978 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1979 * the tunnel_id from val has the vni value
1981 struct ib_flow_spec_tunnel {
1984 struct ib_flow_tunnel_filter val;
1985 struct ib_flow_tunnel_filter mask;
1988 struct ib_flow_esp_filter {
1995 struct ib_flow_spec_esp {
1998 struct ib_flow_esp_filter val;
1999 struct ib_flow_esp_filter mask;
2002 struct ib_flow_gre_filter {
2003 __be16 c_ks_res0_ver;
2010 struct ib_flow_spec_gre {
2013 struct ib_flow_gre_filter val;
2014 struct ib_flow_gre_filter mask;
2017 struct ib_flow_mpls_filter {
2023 struct ib_flow_spec_mpls {
2026 struct ib_flow_mpls_filter val;
2027 struct ib_flow_mpls_filter mask;
2030 struct ib_flow_spec_action_tag {
2031 enum ib_flow_spec_type type;
2036 struct ib_flow_spec_action_drop {
2037 enum ib_flow_spec_type type;
2041 struct ib_flow_spec_action_handle {
2042 enum ib_flow_spec_type type;
2044 struct ib_flow_action *act;
2047 enum ib_counters_description {
2052 struct ib_flow_spec_action_count {
2053 enum ib_flow_spec_type type;
2055 struct ib_counters *counters;
2058 union ib_flow_spec {
2063 struct ib_flow_spec_eth eth;
2064 struct ib_flow_spec_ib ib;
2065 struct ib_flow_spec_ipv4 ipv4;
2066 struct ib_flow_spec_tcp_udp tcp_udp;
2067 struct ib_flow_spec_ipv6 ipv6;
2068 struct ib_flow_spec_tunnel tunnel;
2069 struct ib_flow_spec_esp esp;
2070 struct ib_flow_spec_gre gre;
2071 struct ib_flow_spec_mpls mpls;
2072 struct ib_flow_spec_action_tag flow_tag;
2073 struct ib_flow_spec_action_drop drop;
2074 struct ib_flow_spec_action_handle action;
2075 struct ib_flow_spec_action_count flow_count;
2078 struct ib_flow_attr {
2079 enum ib_flow_attr_type type;
2085 union ib_flow_spec flows[];
2090 struct ib_device *device;
2091 struct ib_uobject *uobject;
2094 enum ib_flow_action_type {
2095 IB_FLOW_ACTION_UNSPECIFIED,
2096 IB_FLOW_ACTION_ESP = 1,
2099 struct ib_flow_action_attrs_esp_keymats {
2100 enum ib_uverbs_flow_action_esp_keymat protocol;
2102 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2106 struct ib_flow_action_attrs_esp_replays {
2107 enum ib_uverbs_flow_action_esp_replay protocol;
2109 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2113 enum ib_flow_action_attrs_esp_flags {
2114 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2115 * This is done in order to share the same flags between user-space and
2116 * kernel and spare an unnecessary translation.
2120 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2121 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2124 struct ib_flow_spec_list {
2125 struct ib_flow_spec_list *next;
2126 union ib_flow_spec spec;
2129 struct ib_flow_action_attrs_esp {
2130 struct ib_flow_action_attrs_esp_keymats *keymat;
2131 struct ib_flow_action_attrs_esp_replays *replay;
2132 struct ib_flow_spec_list *encap;
2133 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2134 * Value of 0 is a valid value.
2140 /* Use enum ib_flow_action_attrs_esp_flags */
2142 u64 hard_limit_pkts;
2145 struct ib_flow_action {
2146 struct ib_device *device;
2147 struct ib_uobject *uobject;
2148 enum ib_flow_action_type type;
2155 enum ib_process_mad_flags {
2156 IB_MAD_IGNORE_MKEY = 1,
2157 IB_MAD_IGNORE_BKEY = 2,
2158 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2161 enum ib_mad_result {
2162 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2163 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2164 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2165 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2168 struct ib_port_cache {
2170 struct ib_pkey_cache *pkey;
2171 struct ib_gid_table *gid;
2173 enum ib_port_state port_state;
2178 struct ib_event_handler event_handler;
2179 struct ib_port_cache *ports;
2184 struct ib_port_immutable {
2191 /* rdma netdev type - specifies protocol type */
2192 enum rdma_netdev_t {
2193 RDMA_NETDEV_OPA_VNIC,
2198 * struct rdma_netdev - rdma netdev
2199 * For cases where netstack interfacing is required.
2201 struct rdma_netdev {
2203 struct ib_device *hca;
2207 * cleanup function must be specified.
2208 * FIXME: This is only used for OPA_VNIC and that usage should be
2211 void (*free_rdma_netdev)(struct net_device *netdev);
2213 /* control functions */
2214 void (*set_id)(struct net_device *netdev, int id);
2216 int (*send)(struct net_device *dev, struct sk_buff *skb,
2217 struct ib_ah *address, u32 dqpn);
2219 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2220 union ib_gid *gid, u16 mlid,
2221 int set_qkey, u32 qkey);
2222 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2223 union ib_gid *gid, u16 mlid);
2226 struct rdma_netdev_alloc_params {
2232 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2233 struct net_device *netdev, void *param);
2236 struct ib_port_pkey_list {
2237 /* Lock to hold while modifying the list. */
2238 spinlock_t list_lock;
2239 struct list_head pkey_list;
2242 struct ib_counters {
2243 struct ib_device *device;
2244 struct ib_uobject *uobject;
2245 /* num of objects attached */
2249 struct ib_counters_read_attr {
2252 u32 flags; /* use enum ib_read_counters_flags */
2255 struct uverbs_attr_bundle;
2258 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2259 struct device *dma_device;
2261 char name[IB_DEVICE_NAME_MAX];
2263 struct list_head event_handler_list;
2264 spinlock_t event_handler_lock;
2266 spinlock_t client_data_lock;
2267 struct list_head core_list;
2268 /* Access to the client_data_list is protected by the client_data_lock
2269 * spinlock and the lists_rwsem read-write semaphore */
2270 struct list_head client_data_list;
2272 struct ib_cache cache;
2274 * port_immutable is indexed by port number
2276 struct ib_port_immutable *port_immutable;
2278 int num_comp_vectors;
2280 struct ib_port_pkey_list *port_pkey_list;
2282 struct iw_cm_verbs *iwcm;
2285 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2286 * driver initialized data. The struct is kfree()'ed by the sysfs
2287 * core when the device is removed. A lifespan of -1 in the return
2288 * struct tells the core to set a default lifespan.
2290 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2293 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2294 * @index - The index in the value array we wish to have updated, or
2295 * num_counters if we want all stats updated
2297 * < 0 - Error, no counters updated
2298 * index - Updated the single counter pointed to by index
2299 * num_counters - Updated all counters (will reset the timestamp
2300 * and prevent further calls for lifespan milliseconds)
2301 * Drivers are allowed to update all counters in leiu of just the
2302 * one given in index at their option
2304 int (*get_hw_stats)(struct ib_device *device,
2305 struct rdma_hw_stats *stats,
2306 u8 port, int index);
2307 int (*query_device)(struct ib_device *device,
2308 struct ib_device_attr *device_attr,
2309 struct ib_udata *udata);
2310 int (*query_port)(struct ib_device *device,
2312 struct ib_port_attr *port_attr);
2313 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2315 /* When calling get_netdev, the HW vendor's driver should return the
2316 * net device of device @device at port @port_num or NULL if such
2317 * a net device doesn't exist. The vendor driver should call dev_hold
2318 * on this net device. The HW vendor's device driver must guarantee
2319 * that this function returns NULL before the net device has finished
2320 * NETDEV_UNREGISTER state.
2322 struct net_device *(*get_netdev)(struct ib_device *device,
2324 /* query_gid should be return GID value for @device, when @port_num
2325 * link layer is either IB or iWarp. It is no-op if @port_num port
2326 * is RoCE link layer.
2328 int (*query_gid)(struct ib_device *device,
2329 u8 port_num, int index,
2331 /* When calling add_gid, the HW vendor's driver should add the gid
2332 * of device of port at gid index available at @attr. Meta-info of
2333 * that gid (for example, the network device related to this gid) is
2334 * available at @attr. @context allows the HW vendor driver to store
2335 * extra information together with a GID entry. The HW vendor driver may
2336 * allocate memory to contain this information and store it in @context
2337 * when a new GID entry is written to. Params are consistent until the
2338 * next call of add_gid or delete_gid. The function should return 0 on
2339 * success or error otherwise. The function could be called
2340 * concurrently for different ports. This function is only called when
2341 * roce_gid_table is used.
2343 int (*add_gid)(const struct ib_gid_attr *attr,
2345 /* When calling del_gid, the HW vendor's driver should delete the
2346 * gid of device @device at gid index gid_index of port port_num
2347 * available in @attr.
2348 * Upon the deletion of a GID entry, the HW vendor must free any
2349 * allocated memory. The caller will clear @context afterwards.
2350 * This function is only called when roce_gid_table is used.
2352 int (*del_gid)(const struct ib_gid_attr *attr,
2354 int (*query_pkey)(struct ib_device *device,
2355 u8 port_num, u16 index, u16 *pkey);
2356 int (*modify_device)(struct ib_device *device,
2357 int device_modify_mask,
2358 struct ib_device_modify *device_modify);
2359 int (*modify_port)(struct ib_device *device,
2360 u8 port_num, int port_modify_mask,
2361 struct ib_port_modify *port_modify);
2362 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2363 struct ib_udata *udata);
2364 int (*dealloc_ucontext)(struct ib_ucontext *context);
2365 int (*mmap)(struct ib_ucontext *context,
2366 struct vm_area_struct *vma);
2367 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2368 struct ib_ucontext *context,
2369 struct ib_udata *udata);
2370 int (*dealloc_pd)(struct ib_pd *pd);
2371 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2372 struct rdma_ah_attr *ah_attr,
2373 struct ib_udata *udata);
2374 int (*modify_ah)(struct ib_ah *ah,
2375 struct rdma_ah_attr *ah_attr);
2376 int (*query_ah)(struct ib_ah *ah,
2377 struct rdma_ah_attr *ah_attr);
2378 int (*destroy_ah)(struct ib_ah *ah);
2379 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2380 struct ib_srq_init_attr *srq_init_attr,
2381 struct ib_udata *udata);
2382 int (*modify_srq)(struct ib_srq *srq,
2383 struct ib_srq_attr *srq_attr,
2384 enum ib_srq_attr_mask srq_attr_mask,
2385 struct ib_udata *udata);
2386 int (*query_srq)(struct ib_srq *srq,
2387 struct ib_srq_attr *srq_attr);
2388 int (*destroy_srq)(struct ib_srq *srq);
2389 int (*post_srq_recv)(struct ib_srq *srq,
2390 const struct ib_recv_wr *recv_wr,
2391 const struct ib_recv_wr **bad_recv_wr);
2392 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2393 struct ib_qp_init_attr *qp_init_attr,
2394 struct ib_udata *udata);
2395 int (*modify_qp)(struct ib_qp *qp,
2396 struct ib_qp_attr *qp_attr,
2398 struct ib_udata *udata);
2399 int (*query_qp)(struct ib_qp *qp,
2400 struct ib_qp_attr *qp_attr,
2402 struct ib_qp_init_attr *qp_init_attr);
2403 int (*destroy_qp)(struct ib_qp *qp);
2404 int (*post_send)(struct ib_qp *qp,
2405 const struct ib_send_wr *send_wr,
2406 const struct ib_send_wr **bad_send_wr);
2407 int (*post_recv)(struct ib_qp *qp,
2408 const struct ib_recv_wr *recv_wr,
2409 const struct ib_recv_wr **bad_recv_wr);
2410 struct ib_cq * (*create_cq)(struct ib_device *device,
2411 const struct ib_cq_init_attr *attr,
2412 struct ib_ucontext *context,
2413 struct ib_udata *udata);
2414 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2416 int (*destroy_cq)(struct ib_cq *cq);
2417 int (*resize_cq)(struct ib_cq *cq, int cqe,
2418 struct ib_udata *udata);
2419 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2421 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2422 int (*req_notify_cq)(struct ib_cq *cq,
2423 enum ib_cq_notify_flags flags);
2424 int (*req_ncomp_notif)(struct ib_cq *cq,
2426 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2427 int mr_access_flags);
2428 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2429 u64 start, u64 length,
2431 int mr_access_flags,
2432 struct ib_udata *udata);
2433 int (*rereg_user_mr)(struct ib_mr *mr,
2435 u64 start, u64 length,
2437 int mr_access_flags,
2439 struct ib_udata *udata);
2440 int (*dereg_mr)(struct ib_mr *mr);
2441 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2442 enum ib_mr_type mr_type,
2444 int (*map_mr_sg)(struct ib_mr *mr,
2445 struct scatterlist *sg,
2447 unsigned int *sg_offset);
2448 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2449 enum ib_mw_type type,
2450 struct ib_udata *udata);
2451 int (*dealloc_mw)(struct ib_mw *mw);
2452 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2453 int mr_access_flags,
2454 struct ib_fmr_attr *fmr_attr);
2455 int (*map_phys_fmr)(struct ib_fmr *fmr,
2456 u64 *page_list, int list_len,
2458 int (*unmap_fmr)(struct list_head *fmr_list);
2459 int (*dealloc_fmr)(struct ib_fmr *fmr);
2460 int (*attach_mcast)(struct ib_qp *qp,
2463 int (*detach_mcast)(struct ib_qp *qp,
2466 int (*process_mad)(struct ib_device *device,
2467 int process_mad_flags,
2469 const struct ib_wc *in_wc,
2470 const struct ib_grh *in_grh,
2471 const struct ib_mad_hdr *in_mad,
2473 struct ib_mad_hdr *out_mad,
2474 size_t *out_mad_size,
2475 u16 *out_mad_pkey_index);
2476 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2477 struct ib_ucontext *ucontext,
2478 struct ib_udata *udata);
2479 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2480 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2484 struct ib_udata *udata);
2485 int (*destroy_flow)(struct ib_flow *flow_id);
2486 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2487 struct ib_mr_status *mr_status);
2488 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2489 void (*drain_rq)(struct ib_qp *qp);
2490 void (*drain_sq)(struct ib_qp *qp);
2491 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2493 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2494 struct ifla_vf_info *ivf);
2495 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2496 struct ifla_vf_stats *stats);
2497 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2499 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2500 struct ib_wq_init_attr *init_attr,
2501 struct ib_udata *udata);
2502 int (*destroy_wq)(struct ib_wq *wq);
2503 int (*modify_wq)(struct ib_wq *wq,
2504 struct ib_wq_attr *attr,
2506 struct ib_udata *udata);
2507 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2508 struct ib_rwq_ind_table_init_attr *init_attr,
2509 struct ib_udata *udata);
2510 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2511 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2512 const struct ib_flow_action_attrs_esp *attr,
2513 struct uverbs_attr_bundle *attrs);
2514 int (*destroy_flow_action)(struct ib_flow_action *action);
2515 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2516 const struct ib_flow_action_attrs_esp *attr,
2517 struct uverbs_attr_bundle *attrs);
2518 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2519 struct ib_ucontext *context,
2520 struct ib_dm_alloc_attr *attr,
2521 struct uverbs_attr_bundle *attrs);
2522 int (*dealloc_dm)(struct ib_dm *dm);
2523 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2524 struct ib_dm_mr_attr *attr,
2525 struct uverbs_attr_bundle *attrs);
2526 struct ib_counters * (*create_counters)(struct ib_device *device,
2527 struct uverbs_attr_bundle *attrs);
2528 int (*destroy_counters)(struct ib_counters *counters);
2529 int (*read_counters)(struct ib_counters *counters,
2530 struct ib_counters_read_attr *counters_read_attr,
2531 struct uverbs_attr_bundle *attrs);
2534 * rdma netdev operation
2536 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2537 * must return -EOPNOTSUPP if it doesn't support the specified type.
2539 struct net_device *(*alloc_rdma_netdev)(
2540 struct ib_device *device,
2542 enum rdma_netdev_t type,
2544 unsigned char name_assign_type,
2545 void (*setup)(struct net_device *));
2547 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2548 enum rdma_netdev_t type,
2549 struct rdma_netdev_alloc_params *params);
2551 struct module *owner;
2553 struct kobject *ports_parent;
2554 struct list_head port_list;
2557 IB_DEV_UNINITIALIZED,
2563 u64 uverbs_cmd_mask;
2564 u64 uverbs_ex_cmd_mask;
2566 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2572 struct ib_device_attr attrs;
2573 struct attribute_group *hw_stats_ag;
2574 struct rdma_hw_stats *hw_stats;
2576 #ifdef CONFIG_CGROUP_RDMA
2577 struct rdmacg_device cg_device;
2582 * Implementation details of the RDMA core, don't use in drivers
2584 struct rdma_restrack_root res;
2587 * The following mandatory functions are used only at device
2588 * registration. Keep functions such as these at the end of this
2589 * structure to avoid cache line misses when accessing struct ib_device
2592 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2593 void (*get_dev_fw_str)(struct ib_device *, char *str);
2594 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2597 const struct uverbs_object_tree_def *const *driver_specs;
2598 enum rdma_driver_id driver_id;
2603 void (*add) (struct ib_device *);
2604 void (*remove)(struct ib_device *, void *client_data);
2606 /* Returns the net_dev belonging to this ib_client and matching the
2608 * @dev: An RDMA device that the net_dev use for communication.
2609 * @port: A physical port number on the RDMA device.
2610 * @pkey: P_Key that the net_dev uses if applicable.
2611 * @gid: A GID that the net_dev uses to communicate.
2612 * @addr: An IP address the net_dev is configured with.
2613 * @client_data: The device's client data set by ib_set_client_data().
2615 * An ib_client that implements a net_dev on top of RDMA devices
2616 * (such as IP over IB) should implement this callback, allowing the
2617 * rdma_cm module to find the right net_dev for a given request.
2619 * The caller is responsible for calling dev_put on the returned
2621 struct net_device *(*get_net_dev_by_params)(
2622 struct ib_device *dev,
2625 const union ib_gid *gid,
2626 const struct sockaddr *addr,
2628 struct list_head list;
2631 struct ib_device *ib_alloc_device(size_t size);
2632 void ib_dealloc_device(struct ib_device *device);
2634 void ib_get_device_fw_str(struct ib_device *device, char *str);
2636 int ib_register_device(struct ib_device *device,
2637 int (*port_callback)(struct ib_device *,
2638 u8, struct kobject *));
2639 void ib_unregister_device(struct ib_device *device);
2641 int ib_register_client (struct ib_client *client);
2642 void ib_unregister_client(struct ib_client *client);
2644 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2645 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2648 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2650 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2653 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2655 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2658 static inline bool ib_is_buffer_cleared(const void __user *p,
2664 if (len > USHRT_MAX)
2667 buf = memdup_user(p, len);
2671 ret = !memchr_inv(buf, 0, len);
2676 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2680 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2684 * ib_is_destroy_retryable - Check whether the uobject destruction
2686 * @ret: The initial destruction return code
2687 * @why: remove reason
2688 * @uobj: The uobject that is destroyed
2690 * This function is a helper function that IB layer and low-level drivers
2691 * can use to consider whether the destruction of the given uobject is
2693 * It checks the original return code, if it wasn't success the destruction
2694 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2695 * the remove reason. (i.e. why).
2696 * Must be called with the object locked for destroy.
2698 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2699 struct ib_uobject *uobj)
2701 return ret && (why == RDMA_REMOVE_DESTROY ||
2702 uobj->context->cleanup_retryable);
2706 * ib_destroy_usecnt - Called during destruction to check the usecnt
2707 * @usecnt: The usecnt atomic
2708 * @why: remove reason
2709 * @uobj: The uobject that is destroyed
2711 * Non-zero usecnts will block destruction unless destruction was triggered by
2712 * a ucontext cleanup.
2714 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2715 enum rdma_remove_reason why,
2716 struct ib_uobject *uobj)
2718 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2724 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2725 * contains all required attributes and no attributes not allowed for
2726 * the given QP state transition.
2727 * @cur_state: Current QP state
2728 * @next_state: Next QP state
2730 * @mask: Mask of supplied QP attributes
2731 * @ll : link layer of port
2733 * This function is a helper function that a low-level driver's
2734 * modify_qp method can use to validate the consumer's input. It
2735 * checks that cur_state and next_state are valid QP states, that a
2736 * transition from cur_state to next_state is allowed by the IB spec,
2737 * and that the attribute mask supplied is allowed for the transition.
2739 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2740 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2741 enum rdma_link_layer ll);
2743 void ib_register_event_handler(struct ib_event_handler *event_handler);
2744 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2745 void ib_dispatch_event(struct ib_event *event);
2747 int ib_query_port(struct ib_device *device,
2748 u8 port_num, struct ib_port_attr *port_attr);
2750 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2754 * rdma_cap_ib_switch - Check if the device is IB switch
2755 * @device: Device to check
2757 * Device driver is responsible for setting is_switch bit on
2758 * in ib_device structure at init time.
2760 * Return: true if the device is IB switch.
2762 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2764 return device->is_switch;
2768 * rdma_start_port - Return the first valid port number for the device
2771 * @device: Device to be checked
2773 * Return start port number
2775 static inline u8 rdma_start_port(const struct ib_device *device)
2777 return rdma_cap_ib_switch(device) ? 0 : 1;
2781 * rdma_end_port - Return the last valid port number for the device
2784 * @device: Device to be checked
2786 * Return last port number
2788 static inline u8 rdma_end_port(const struct ib_device *device)
2790 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2793 static inline int rdma_is_port_valid(const struct ib_device *device,
2796 return (port >= rdma_start_port(device) &&
2797 port <= rdma_end_port(device));
2800 static inline bool rdma_is_grh_required(const struct ib_device *device,
2803 return device->port_immutable[port_num].core_cap_flags &
2804 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2807 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2809 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2812 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2814 return device->port_immutable[port_num].core_cap_flags &
2815 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2818 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2823 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2825 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2828 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2830 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2833 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2835 return rdma_protocol_ib(device, port_num) ||
2836 rdma_protocol_roce(device, port_num);
2839 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2841 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2844 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2846 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2850 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2851 * Management Datagrams.
2852 * @device: Device to check
2853 * @port_num: Port number to check
2855 * Management Datagrams (MAD) are a required part of the InfiniBand
2856 * specification and are supported on all InfiniBand devices. A slightly
2857 * extended version are also supported on OPA interfaces.
2859 * Return: true if the port supports sending/receiving of MAD packets.
2861 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2863 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2867 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2868 * Management Datagrams.
2869 * @device: Device to check
2870 * @port_num: Port number to check
2872 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2873 * datagrams with their own versions. These OPA MADs share many but not all of
2874 * the characteristics of InfiniBand MADs.
2876 * OPA MADs differ in the following ways:
2878 * 1) MADs are variable size up to 2K
2879 * IBTA defined MADs remain fixed at 256 bytes
2880 * 2) OPA SMPs must carry valid PKeys
2881 * 3) OPA SMP packets are a different format
2883 * Return: true if the port supports OPA MAD packet formats.
2885 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2887 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2888 == RDMA_CORE_CAP_OPA_MAD;
2892 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2893 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2894 * @device: Device to check
2895 * @port_num: Port number to check
2897 * Each InfiniBand node is required to provide a Subnet Management Agent
2898 * that the subnet manager can access. Prior to the fabric being fully
2899 * configured by the subnet manager, the SMA is accessed via a well known
2900 * interface called the Subnet Management Interface (SMI). This interface
2901 * uses directed route packets to communicate with the SM to get around the
2902 * chicken and egg problem of the SM needing to know what's on the fabric
2903 * in order to configure the fabric, and needing to configure the fabric in
2904 * order to send packets to the devices on the fabric. These directed
2905 * route packets do not need the fabric fully configured in order to reach
2906 * their destination. The SMI is the only method allowed to send
2907 * directed route packets on an InfiniBand fabric.
2909 * Return: true if the port provides an SMI.
2911 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2913 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2917 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2918 * Communication Manager.
2919 * @device: Device to check
2920 * @port_num: Port number to check
2922 * The InfiniBand Communication Manager is one of many pre-defined General
2923 * Service Agents (GSA) that are accessed via the General Service
2924 * Interface (GSI). It's role is to facilitate establishment of connections
2925 * between nodes as well as other management related tasks for established
2928 * Return: true if the port supports an IB CM (this does not guarantee that
2929 * a CM is actually running however).
2931 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2933 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2937 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2938 * Communication Manager.
2939 * @device: Device to check
2940 * @port_num: Port number to check
2942 * Similar to above, but specific to iWARP connections which have a different
2943 * managment protocol than InfiniBand.
2945 * Return: true if the port supports an iWARP CM (this does not guarantee that
2946 * a CM is actually running however).
2948 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2950 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2954 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2955 * Subnet Administration.
2956 * @device: Device to check
2957 * @port_num: Port number to check
2959 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2960 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2961 * fabrics, devices should resolve routes to other hosts by contacting the
2962 * SA to query the proper route.
2964 * Return: true if the port should act as a client to the fabric Subnet
2965 * Administration interface. This does not imply that the SA service is
2968 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2970 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2974 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2976 * @device: Device to check
2977 * @port_num: Port number to check
2979 * InfiniBand multicast registration is more complex than normal IPv4 or
2980 * IPv6 multicast registration. Each Host Channel Adapter must register
2981 * with the Subnet Manager when it wishes to join a multicast group. It
2982 * should do so only once regardless of how many queue pairs it subscribes
2983 * to this group. And it should leave the group only after all queue pairs
2984 * attached to the group have been detached.
2986 * Return: true if the port must undertake the additional adminstrative
2987 * overhead of registering/unregistering with the SM and tracking of the
2988 * total number of queue pairs attached to the multicast group.
2990 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2992 return rdma_cap_ib_sa(device, port_num);
2996 * rdma_cap_af_ib - Check if the port of device has the capability
2997 * Native Infiniband Address.
2998 * @device: Device to check
2999 * @port_num: Port number to check
3001 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3002 * GID. RoCE uses a different mechanism, but still generates a GID via
3003 * a prescribed mechanism and port specific data.
3005 * Return: true if the port uses a GID address to identify devices on the
3008 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3010 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3014 * rdma_cap_eth_ah - Check if the port of device has the capability
3015 * Ethernet Address Handle.
3016 * @device: Device to check
3017 * @port_num: Port number to check
3019 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3020 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3021 * port. Normally, packet headers are generated by the sending host
3022 * adapter, but when sending connectionless datagrams, we must manually
3023 * inject the proper headers for the fabric we are communicating over.
3025 * Return: true if we are running as a RoCE port and must force the
3026 * addition of a Global Route Header built from our Ethernet Address
3027 * Handle into our header list for connectionless packets.
3029 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3031 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3035 * rdma_cap_opa_ah - Check if the port of device supports
3036 * OPA Address handles
3037 * @device: Device to check
3038 * @port_num: Port number to check
3040 * Return: true if we are running on an OPA device which supports
3041 * the extended OPA addressing.
3043 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3045 return (device->port_immutable[port_num].core_cap_flags &
3046 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3050 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3053 * @port_num: Port number
3055 * This MAD size includes the MAD headers and MAD payload. No other headers
3058 * Return the max MAD size required by the Port. Will return 0 if the port
3059 * does not support MADs
3061 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3063 return device->port_immutable[port_num].max_mad_size;
3067 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3068 * @device: Device to check
3069 * @port_num: Port number to check
3071 * RoCE GID table mechanism manages the various GIDs for a device.
3073 * NOTE: if allocating the port's GID table has failed, this call will still
3074 * return true, but any RoCE GID table API will fail.
3076 * Return: true if the port uses RoCE GID table mechanism in order to manage
3079 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3082 return rdma_protocol_roce(device, port_num) &&
3083 device->add_gid && device->del_gid;
3087 * Check if the device supports READ W/ INVALIDATE.
3089 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3092 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3093 * has support for it yet.
3095 return rdma_protocol_iwarp(dev, port_num);
3098 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3100 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3101 struct ifla_vf_info *info);
3102 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3103 struct ifla_vf_stats *stats);
3104 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3107 int ib_query_pkey(struct ib_device *device,
3108 u8 port_num, u16 index, u16 *pkey);
3110 int ib_modify_device(struct ib_device *device,
3111 int device_modify_mask,
3112 struct ib_device_modify *device_modify);
3114 int ib_modify_port(struct ib_device *device,
3115 u8 port_num, int port_modify_mask,
3116 struct ib_port_modify *port_modify);
3118 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3119 u8 *port_num, u16 *index);
3121 int ib_find_pkey(struct ib_device *device,
3122 u8 port_num, u16 pkey, u16 *index);
3126 * Create a memory registration for all memory in the system and place
3127 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3128 * ULPs to avoid the overhead of dynamic MRs.
3130 * This flag is generally considered unsafe and must only be used in
3131 * extremly trusted environments. Every use of it will log a warning
3132 * in the kernel log.
3134 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3137 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3138 const char *caller);
3139 #define ib_alloc_pd(device, flags) \
3140 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3141 void ib_dealloc_pd(struct ib_pd *pd);
3144 * rdma_create_ah - Creates an address handle for the given address vector.
3145 * @pd: The protection domain associated with the address handle.
3146 * @ah_attr: The attributes of the address vector.
3148 * The address handle is used to reference a local or global destination
3149 * in all UD QP post sends.
3151 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3154 * rdma_create_user_ah - Creates an address handle for the given address vector.
3155 * It resolves destination mac address for ah attribute of RoCE type.
3156 * @pd: The protection domain associated with the address handle.
3157 * @ah_attr: The attributes of the address vector.
3158 * @udata: pointer to user's input output buffer information need by
3161 * It returns 0 on success and returns appropriate error code on error.
3162 * The address handle is used to reference a local or global destination
3163 * in all UD QP post sends.
3165 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3166 struct rdma_ah_attr *ah_attr,
3167 struct ib_udata *udata);
3169 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3171 * @hdr: the L3 header to parse
3172 * @net_type: type of header to parse
3173 * @sgid: place to store source gid
3174 * @dgid: place to store destination gid
3176 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3177 enum rdma_network_type net_type,
3178 union ib_gid *sgid, union ib_gid *dgid);
3181 * ib_get_rdma_header_version - Get the header version
3182 * @hdr: the L3 header to parse
3184 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3187 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3189 * @device: Device on which the received message arrived.
3190 * @port_num: Port on which the received message arrived.
3191 * @wc: Work completion associated with the received message.
3192 * @grh: References the received global route header. This parameter is
3193 * ignored unless the work completion indicates that the GRH is valid.
3194 * @ah_attr: Returned attributes that can be used when creating an address
3195 * handle for replying to the message.
3196 * When ib_init_ah_attr_from_wc() returns success,
3197 * (a) for IB link layer it optionally contains a reference to SGID attribute
3198 * when GRH is present for IB link layer.
3199 * (b) for RoCE link layer it contains a reference to SGID attribute.
3200 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3201 * attributes which are initialized using ib_init_ah_attr_from_wc().
3204 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3205 const struct ib_wc *wc, const struct ib_grh *grh,
3206 struct rdma_ah_attr *ah_attr);
3209 * ib_create_ah_from_wc - Creates an address handle associated with the
3210 * sender of the specified work completion.
3211 * @pd: The protection domain associated with the address handle.
3212 * @wc: Work completion information associated with a received message.
3213 * @grh: References the received global route header. This parameter is
3214 * ignored unless the work completion indicates that the GRH is valid.
3215 * @port_num: The outbound port number to associate with the address.
3217 * The address handle is used to reference a local or global destination
3218 * in all UD QP post sends.
3220 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3221 const struct ib_grh *grh, u8 port_num);
3224 * rdma_modify_ah - Modifies the address vector associated with an address
3226 * @ah: The address handle to modify.
3227 * @ah_attr: The new address vector attributes to associate with the
3230 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3233 * rdma_query_ah - Queries the address vector associated with an address
3235 * @ah: The address handle to query.
3236 * @ah_attr: The address vector attributes associated with the address
3239 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3242 * rdma_destroy_ah - Destroys an address handle.
3243 * @ah: The address handle to destroy.
3245 int rdma_destroy_ah(struct ib_ah *ah);
3248 * ib_create_srq - Creates a SRQ associated with the specified protection
3250 * @pd: The protection domain associated with the SRQ.
3251 * @srq_init_attr: A list of initial attributes required to create the
3252 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3253 * the actual capabilities of the created SRQ.
3255 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3256 * requested size of the SRQ, and set to the actual values allocated
3257 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3258 * will always be at least as large as the requested values.
3260 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3261 struct ib_srq_init_attr *srq_init_attr);
3264 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3265 * @srq: The SRQ to modify.
3266 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3267 * the current values of selected SRQ attributes are returned.
3268 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3269 * are being modified.
3271 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3272 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3273 * the number of receives queued drops below the limit.
3275 int ib_modify_srq(struct ib_srq *srq,
3276 struct ib_srq_attr *srq_attr,
3277 enum ib_srq_attr_mask srq_attr_mask);
3280 * ib_query_srq - Returns the attribute list and current values for the
3282 * @srq: The SRQ to query.
3283 * @srq_attr: The attributes of the specified SRQ.
3285 int ib_query_srq(struct ib_srq *srq,
3286 struct ib_srq_attr *srq_attr);
3289 * ib_destroy_srq - Destroys the specified SRQ.
3290 * @srq: The SRQ to destroy.
3292 int ib_destroy_srq(struct ib_srq *srq);
3295 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3296 * @srq: The SRQ to post the work request on.
3297 * @recv_wr: A list of work requests to post on the receive queue.
3298 * @bad_recv_wr: On an immediate failure, this parameter will reference
3299 * the work request that failed to be posted on the QP.
3301 static inline int ib_post_srq_recv(struct ib_srq *srq,
3302 const struct ib_recv_wr *recv_wr,
3303 const struct ib_recv_wr **bad_recv_wr)
3305 const struct ib_recv_wr *dummy;
3307 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr ? : &dummy);
3311 * ib_create_qp - Creates a QP associated with the specified protection
3313 * @pd: The protection domain associated with the QP.
3314 * @qp_init_attr: A list of initial attributes required to create the
3315 * QP. If QP creation succeeds, then the attributes are updated to
3316 * the actual capabilities of the created QP.
3318 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3319 struct ib_qp_init_attr *qp_init_attr);
3322 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3323 * @qp: The QP to modify.
3324 * @attr: On input, specifies the QP attributes to modify. On output,
3325 * the current values of selected QP attributes are returned.
3326 * @attr_mask: A bit-mask used to specify which attributes of the QP
3327 * are being modified.
3328 * @udata: pointer to user's input output buffer information
3329 * are being modified.
3330 * It returns 0 on success and returns appropriate error code on error.
3332 int ib_modify_qp_with_udata(struct ib_qp *qp,
3333 struct ib_qp_attr *attr,
3335 struct ib_udata *udata);
3338 * ib_modify_qp - Modifies the attributes for the specified QP and then
3339 * transitions the QP to the given state.
3340 * @qp: The QP to modify.
3341 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3342 * the current values of selected QP attributes are returned.
3343 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3344 * are being modified.
3346 int ib_modify_qp(struct ib_qp *qp,
3347 struct ib_qp_attr *qp_attr,
3351 * ib_query_qp - Returns the attribute list and current values for the
3353 * @qp: The QP to query.
3354 * @qp_attr: The attributes of the specified QP.
3355 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3356 * @qp_init_attr: Additional attributes of the selected QP.
3358 * The qp_attr_mask may be used to limit the query to gathering only the
3359 * selected attributes.
3361 int ib_query_qp(struct ib_qp *qp,
3362 struct ib_qp_attr *qp_attr,
3364 struct ib_qp_init_attr *qp_init_attr);
3367 * ib_destroy_qp - Destroys the specified QP.
3368 * @qp: The QP to destroy.
3370 int ib_destroy_qp(struct ib_qp *qp);
3373 * ib_open_qp - Obtain a reference to an existing sharable QP.
3374 * @xrcd - XRC domain
3375 * @qp_open_attr: Attributes identifying the QP to open.
3377 * Returns a reference to a sharable QP.
3379 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3380 struct ib_qp_open_attr *qp_open_attr);
3383 * ib_close_qp - Release an external reference to a QP.
3384 * @qp: The QP handle to release
3386 * The opened QP handle is released by the caller. The underlying
3387 * shared QP is not destroyed until all internal references are released.
3389 int ib_close_qp(struct ib_qp *qp);
3392 * ib_post_send - Posts a list of work requests to the send queue of
3394 * @qp: The QP to post the work request on.
3395 * @send_wr: A list of work requests to post on the send queue.
3396 * @bad_send_wr: On an immediate failure, this parameter will reference
3397 * the work request that failed to be posted on the QP.
3399 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3400 * error is returned, the QP state shall not be affected,
3401 * ib_post_send() will return an immediate error after queueing any
3402 * earlier work requests in the list.
3404 static inline int ib_post_send(struct ib_qp *qp,
3405 const struct ib_send_wr *send_wr,
3406 const struct ib_send_wr **bad_send_wr)
3408 const struct ib_send_wr *dummy;
3410 return qp->device->post_send(qp, send_wr, bad_send_wr ? : &dummy);
3414 * ib_post_recv - Posts a list of work requests to the receive queue of
3416 * @qp: The QP to post the work request on.
3417 * @recv_wr: A list of work requests to post on the receive queue.
3418 * @bad_recv_wr: On an immediate failure, this parameter will reference
3419 * the work request that failed to be posted on the QP.
3421 static inline int ib_post_recv(struct ib_qp *qp,
3422 const struct ib_recv_wr *recv_wr,
3423 const struct ib_recv_wr **bad_recv_wr)
3425 const struct ib_recv_wr *dummy;
3427 return qp->device->post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3430 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3431 int nr_cqe, int comp_vector,
3432 enum ib_poll_context poll_ctx, const char *caller);
3433 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3434 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3436 void ib_free_cq(struct ib_cq *cq);
3437 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3440 * ib_create_cq - Creates a CQ on the specified device.
3441 * @device: The device on which to create the CQ.
3442 * @comp_handler: A user-specified callback that is invoked when a
3443 * completion event occurs on the CQ.
3444 * @event_handler: A user-specified callback that is invoked when an
3445 * asynchronous event not associated with a completion occurs on the CQ.
3446 * @cq_context: Context associated with the CQ returned to the user via
3447 * the associated completion and event handlers.
3448 * @cq_attr: The attributes the CQ should be created upon.
3450 * Users can examine the cq structure to determine the actual CQ size.
3452 struct ib_cq *__ib_create_cq(struct ib_device *device,
3453 ib_comp_handler comp_handler,
3454 void (*event_handler)(struct ib_event *, void *),
3456 const struct ib_cq_init_attr *cq_attr,
3457 const char *caller);
3458 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3459 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3462 * ib_resize_cq - Modifies the capacity of the CQ.
3463 * @cq: The CQ to resize.
3464 * @cqe: The minimum size of the CQ.
3466 * Users can examine the cq structure to determine the actual CQ size.
3468 int ib_resize_cq(struct ib_cq *cq, int cqe);
3471 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3472 * @cq: The CQ to modify.
3473 * @cq_count: number of CQEs that will trigger an event
3474 * @cq_period: max period of time in usec before triggering an event
3477 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3480 * ib_destroy_cq - Destroys the specified CQ.
3481 * @cq: The CQ to destroy.
3483 int ib_destroy_cq(struct ib_cq *cq);
3486 * ib_poll_cq - poll a CQ for completion(s)
3487 * @cq:the CQ being polled
3488 * @num_entries:maximum number of completions to return
3489 * @wc:array of at least @num_entries &struct ib_wc where completions
3492 * Poll a CQ for (possibly multiple) completions. If the return value
3493 * is < 0, an error occurred. If the return value is >= 0, it is the
3494 * number of completions returned. If the return value is
3495 * non-negative and < num_entries, then the CQ was emptied.
3497 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3500 return cq->device->poll_cq(cq, num_entries, wc);
3504 * ib_req_notify_cq - Request completion notification on a CQ.
3505 * @cq: The CQ to generate an event for.
3507 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3508 * to request an event on the next solicited event or next work
3509 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3510 * may also be |ed in to request a hint about missed events, as
3514 * < 0 means an error occurred while requesting notification
3515 * == 0 means notification was requested successfully, and if
3516 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3517 * were missed and it is safe to wait for another event. In
3518 * this case is it guaranteed that any work completions added
3519 * to the CQ since the last CQ poll will trigger a completion
3520 * notification event.
3521 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3522 * in. It means that the consumer must poll the CQ again to
3523 * make sure it is empty to avoid missing an event because of a
3524 * race between requesting notification and an entry being
3525 * added to the CQ. This return value means it is possible
3526 * (but not guaranteed) that a work completion has been added
3527 * to the CQ since the last poll without triggering a
3528 * completion notification event.
3530 static inline int ib_req_notify_cq(struct ib_cq *cq,
3531 enum ib_cq_notify_flags flags)
3533 return cq->device->req_notify_cq(cq, flags);
3537 * ib_req_ncomp_notif - Request completion notification when there are
3538 * at least the specified number of unreaped completions on the CQ.
3539 * @cq: The CQ to generate an event for.
3540 * @wc_cnt: The number of unreaped completions that should be on the
3541 * CQ before an event is generated.
3543 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3545 return cq->device->req_ncomp_notif ?
3546 cq->device->req_ncomp_notif(cq, wc_cnt) :
3551 * ib_dma_mapping_error - check a DMA addr for error
3552 * @dev: The device for which the dma_addr was created
3553 * @dma_addr: The DMA address to check
3555 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3557 return dma_mapping_error(dev->dma_device, dma_addr);
3561 * ib_dma_map_single - Map a kernel virtual address to DMA address
3562 * @dev: The device for which the dma_addr is to be created
3563 * @cpu_addr: The kernel virtual address
3564 * @size: The size of the region in bytes
3565 * @direction: The direction of the DMA
3567 static inline u64 ib_dma_map_single(struct ib_device *dev,
3568 void *cpu_addr, size_t size,
3569 enum dma_data_direction direction)
3571 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3575 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3576 * @dev: The device for which the DMA address was created
3577 * @addr: The DMA address
3578 * @size: The size of the region in bytes
3579 * @direction: The direction of the DMA
3581 static inline void ib_dma_unmap_single(struct ib_device *dev,
3582 u64 addr, size_t size,
3583 enum dma_data_direction direction)
3585 dma_unmap_single(dev->dma_device, addr, size, direction);
3589 * ib_dma_map_page - Map a physical page to DMA address
3590 * @dev: The device for which the dma_addr is to be created
3591 * @page: The page to be mapped
3592 * @offset: The offset within the page
3593 * @size: The size of the region in bytes
3594 * @direction: The direction of the DMA
3596 static inline u64 ib_dma_map_page(struct ib_device *dev,
3598 unsigned long offset,
3600 enum dma_data_direction direction)
3602 return dma_map_page(dev->dma_device, page, offset, size, direction);
3606 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3607 * @dev: The device for which the DMA address was created
3608 * @addr: The DMA address
3609 * @size: The size of the region in bytes
3610 * @direction: The direction of the DMA
3612 static inline void ib_dma_unmap_page(struct ib_device *dev,
3613 u64 addr, size_t size,
3614 enum dma_data_direction direction)
3616 dma_unmap_page(dev->dma_device, addr, size, direction);
3620 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3621 * @dev: The device for which the DMA addresses are to be created
3622 * @sg: The array of scatter/gather entries
3623 * @nents: The number of scatter/gather entries
3624 * @direction: The direction of the DMA
3626 static inline int ib_dma_map_sg(struct ib_device *dev,
3627 struct scatterlist *sg, int nents,
3628 enum dma_data_direction direction)
3630 return dma_map_sg(dev->dma_device, sg, nents, direction);
3634 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3635 * @dev: The device for which the DMA addresses were created
3636 * @sg: The array of scatter/gather entries
3637 * @nents: The number of scatter/gather entries
3638 * @direction: The direction of the DMA
3640 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3641 struct scatterlist *sg, int nents,
3642 enum dma_data_direction direction)
3644 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3647 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3648 struct scatterlist *sg, int nents,
3649 enum dma_data_direction direction,
3650 unsigned long dma_attrs)
3652 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3656 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3657 struct scatterlist *sg, int nents,
3658 enum dma_data_direction direction,
3659 unsigned long dma_attrs)
3661 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3664 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3665 * @dev: The device for which the DMA addresses were created
3666 * @sg: The scatter/gather entry
3668 * Note: this function is obsolete. To do: change all occurrences of
3669 * ib_sg_dma_address() into sg_dma_address().
3671 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3672 struct scatterlist *sg)
3674 return sg_dma_address(sg);
3678 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3679 * @dev: The device for which the DMA addresses were created
3680 * @sg: The scatter/gather entry
3682 * Note: this function is obsolete. To do: change all occurrences of
3683 * ib_sg_dma_len() into sg_dma_len().
3685 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3686 struct scatterlist *sg)
3688 return sg_dma_len(sg);
3692 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3693 * @dev: The device for which the DMA address was created
3694 * @addr: The DMA address
3695 * @size: The size of the region in bytes
3696 * @dir: The direction of the DMA
3698 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3701 enum dma_data_direction dir)
3703 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3707 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3708 * @dev: The device for which the DMA address was created
3709 * @addr: The DMA address
3710 * @size: The size of the region in bytes
3711 * @dir: The direction of the DMA
3713 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3716 enum dma_data_direction dir)
3718 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3722 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3723 * @dev: The device for which the DMA address is requested
3724 * @size: The size of the region to allocate in bytes
3725 * @dma_handle: A pointer for returning the DMA address of the region
3726 * @flag: memory allocator flags
3728 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3730 dma_addr_t *dma_handle,
3733 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3737 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3738 * @dev: The device for which the DMA addresses were allocated
3739 * @size: The size of the region
3740 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3741 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3743 static inline void ib_dma_free_coherent(struct ib_device *dev,
3744 size_t size, void *cpu_addr,
3745 dma_addr_t dma_handle)
3747 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3751 * ib_dereg_mr - Deregisters a memory region and removes it from the
3752 * HCA translation table.
3753 * @mr: The memory region to deregister.
3755 * This function can fail, if the memory region has memory windows bound to it.
3757 int ib_dereg_mr(struct ib_mr *mr);
3759 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3760 enum ib_mr_type mr_type,
3764 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3766 * @mr - struct ib_mr pointer to be updated.
3767 * @newkey - new key to be used.
3769 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3771 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3772 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3776 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3777 * for calculating a new rkey for type 2 memory windows.
3778 * @rkey - the rkey to increment.
3780 static inline u32 ib_inc_rkey(u32 rkey)
3782 const u32 mask = 0x000000ff;
3783 return ((rkey + 1) & mask) | (rkey & ~mask);
3787 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3788 * @pd: The protection domain associated with the unmapped region.
3789 * @mr_access_flags: Specifies the memory access rights.
3790 * @fmr_attr: Attributes of the unmapped region.
3792 * A fast memory region must be mapped before it can be used as part of
3795 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3796 int mr_access_flags,
3797 struct ib_fmr_attr *fmr_attr);
3800 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3801 * @fmr: The fast memory region to associate with the pages.
3802 * @page_list: An array of physical pages to map to the fast memory region.
3803 * @list_len: The number of pages in page_list.
3804 * @iova: The I/O virtual address to use with the mapped region.
3806 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3807 u64 *page_list, int list_len,
3810 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3814 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3815 * @fmr_list: A linked list of fast memory regions to unmap.
3817 int ib_unmap_fmr(struct list_head *fmr_list);
3820 * ib_dealloc_fmr - Deallocates a fast memory region.
3821 * @fmr: The fast memory region to deallocate.
3823 int ib_dealloc_fmr(struct ib_fmr *fmr);
3826 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3827 * @qp: QP to attach to the multicast group. The QP must be type
3829 * @gid: Multicast group GID.
3830 * @lid: Multicast group LID in host byte order.
3832 * In order to send and receive multicast packets, subnet
3833 * administration must have created the multicast group and configured
3834 * the fabric appropriately. The port associated with the specified
3835 * QP must also be a member of the multicast group.
3837 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3840 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3841 * @qp: QP to detach from the multicast group.
3842 * @gid: Multicast group GID.
3843 * @lid: Multicast group LID in host byte order.
3845 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3848 * ib_alloc_xrcd - Allocates an XRC domain.
3849 * @device: The device on which to allocate the XRC domain.
3850 * @caller: Module name for kernel consumers
3852 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3853 #define ib_alloc_xrcd(device) \
3854 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3857 * ib_dealloc_xrcd - Deallocates an XRC domain.
3858 * @xrcd: The XRC domain to deallocate.
3860 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3862 static inline int ib_check_mr_access(int flags)
3865 * Local write permission is required if remote write or
3866 * remote atomic permission is also requested.
3868 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3869 !(flags & IB_ACCESS_LOCAL_WRITE))
3875 static inline bool ib_access_writable(int access_flags)
3878 * We have writable memory backing the MR if any of the following
3879 * access flags are set. "Local write" and "remote write" obviously
3880 * require write access. "Remote atomic" can do things like fetch and
3881 * add, which will modify memory, and "MW bind" can change permissions
3882 * by binding a window.
3884 return access_flags &
3885 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3886 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3890 * ib_check_mr_status: lightweight check of MR status.
3891 * This routine may provide status checks on a selected
3892 * ib_mr. first use is for signature status check.
3894 * @mr: A memory region.
3895 * @check_mask: Bitmask of which checks to perform from
3896 * ib_mr_status_check enumeration.
3897 * @mr_status: The container of relevant status checks.
3898 * failed checks will be indicated in the status bitmask
3899 * and the relevant info shall be in the error item.
3901 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3902 struct ib_mr_status *mr_status);
3904 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3905 u16 pkey, const union ib_gid *gid,
3906 const struct sockaddr *addr);
3907 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3908 struct ib_wq_init_attr *init_attr);
3909 int ib_destroy_wq(struct ib_wq *wq);
3910 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3912 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3913 struct ib_rwq_ind_table_init_attr*
3914 wq_ind_table_init_attr);
3915 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3917 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3918 unsigned int *sg_offset, unsigned int page_size);
3921 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3922 unsigned int *sg_offset, unsigned int page_size)
3926 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3932 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3933 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3935 void ib_drain_rq(struct ib_qp *qp);
3936 void ib_drain_sq(struct ib_qp *qp);
3937 void ib_drain_qp(struct ib_qp *qp);
3939 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3941 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3943 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3944 return attr->roce.dmac;
3948 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3950 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3951 attr->ib.dlid = (u16)dlid;
3952 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3953 attr->opa.dlid = dlid;
3956 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3958 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3959 return attr->ib.dlid;
3960 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3961 return attr->opa.dlid;
3965 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3970 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3975 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3978 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3979 attr->ib.src_path_bits = src_path_bits;
3980 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3981 attr->opa.src_path_bits = src_path_bits;
3984 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3986 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3987 return attr->ib.src_path_bits;
3988 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3989 return attr->opa.src_path_bits;
3993 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3996 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3997 attr->opa.make_grd = make_grd;
4000 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4002 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4003 return attr->opa.make_grd;
4007 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4009 attr->port_num = port_num;
4012 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4014 return attr->port_num;
4017 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4020 attr->static_rate = static_rate;
4023 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4025 return attr->static_rate;
4028 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4029 enum ib_ah_flags flag)
4031 attr->ah_flags = flag;
4034 static inline enum ib_ah_flags
4035 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4037 return attr->ah_flags;
4040 static inline const struct ib_global_route
4041 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4046 /*To retrieve and modify the grh */
4047 static inline struct ib_global_route
4048 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4053 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4055 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4057 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4060 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4063 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4065 grh->dgid.global.subnet_prefix = prefix;
4068 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4071 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4073 grh->dgid.global.interface_id = if_id;
4076 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4077 union ib_gid *dgid, u32 flow_label,
4078 u8 sgid_index, u8 hop_limit,
4081 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4083 attr->ah_flags = IB_AH_GRH;
4086 grh->flow_label = flow_label;
4087 grh->sgid_index = sgid_index;
4088 grh->hop_limit = hop_limit;
4089 grh->traffic_class = traffic_class;
4090 grh->sgid_attr = NULL;
4093 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4094 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4095 u32 flow_label, u8 hop_limit, u8 traffic_class,
4096 const struct ib_gid_attr *sgid_attr);
4097 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4098 const struct rdma_ah_attr *src);
4099 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4100 const struct rdma_ah_attr *new);
4101 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4104 * rdma_ah_find_type - Return address handle type.
4106 * @dev: Device to be checked
4107 * @port_num: Port number
4109 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4112 if (rdma_protocol_roce(dev, port_num))
4113 return RDMA_AH_ATTR_TYPE_ROCE;
4114 if (rdma_protocol_ib(dev, port_num)) {
4115 if (rdma_cap_opa_ah(dev, port_num))
4116 return RDMA_AH_ATTR_TYPE_OPA;
4117 return RDMA_AH_ATTR_TYPE_IB;
4120 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4124 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4125 * In the current implementation the only way to get
4126 * get the 32bit lid is from other sources for OPA.
4127 * For IB, lids will always be 16bits so cast the
4128 * value accordingly.
4132 static inline u16 ib_lid_cpu16(u32 lid)
4134 WARN_ON_ONCE(lid & 0xFFFF0000);
4139 * ib_lid_be16 - Return lid in 16bit BE encoding.
4143 static inline __be16 ib_lid_be16(u32 lid)
4145 WARN_ON_ONCE(lid & 0xFFFF0000);
4146 return cpu_to_be16((u16)lid);
4150 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4152 * @device: the rdma device
4153 * @comp_vector: index of completion vector
4155 * Returns NULL on failure, otherwise a corresponding cpu map of the
4156 * completion vector (returns all-cpus map if the device driver doesn't
4157 * implement get_vector_affinity).
4159 static inline const struct cpumask *
4160 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4162 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4163 !device->get_vector_affinity)
4166 return device->get_vector_affinity(device, comp_vector);
4170 static inline void ib_set_flow(struct ib_uobject *uobj, struct ib_flow *ibflow,
4171 struct ib_qp *qp, struct ib_device *device)
4173 uobj->object = ibflow;
4174 ibflow->uobject = uobj;
4177 atomic_inc(&qp->usecnt);
4181 ibflow->device = device;
4185 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4186 * and add their gids, as needed, to the relevant RoCE devices.
4188 * @device: the rdma device
4190 void rdma_roce_rescan_device(struct ib_device *ibdev);
4192 struct ib_ucontext *ib_uverbs_get_ucontext(struct ib_uverbs_file *ufile);
4194 int uverbs_destroy_def_handler(struct ib_uverbs_file *file,
4195 struct uverbs_attr_bundle *attrs);
4197 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4198 enum rdma_netdev_t type, const char *name,
4199 unsigned char name_assign_type,
4200 void (*setup)(struct net_device *));
4202 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4203 enum rdma_netdev_t type, const char *name,
4204 unsigned char name_assign_type,
4205 void (*setup)(struct net_device *),
4206 struct net_device *netdev);
4208 #endif /* IB_VERBS_H */