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
533 /* Protocol 0xFFF00000 */
534 #define RDMA_CORE_CAP_PROT_IB 0x00100000
535 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
536 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
537 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
538 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
539 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
541 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
542 | RDMA_CORE_CAP_IB_MAD \
543 | RDMA_CORE_CAP_IB_SMI \
544 | RDMA_CORE_CAP_IB_CM \
545 | RDMA_CORE_CAP_IB_SA \
546 | RDMA_CORE_CAP_AF_IB)
547 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
548 | RDMA_CORE_CAP_IB_MAD \
549 | RDMA_CORE_CAP_IB_CM \
550 | RDMA_CORE_CAP_AF_IB \
551 | RDMA_CORE_CAP_ETH_AH)
552 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
553 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
554 | RDMA_CORE_CAP_IB_MAD \
555 | RDMA_CORE_CAP_IB_CM \
556 | RDMA_CORE_CAP_AF_IB \
557 | RDMA_CORE_CAP_ETH_AH)
558 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
559 | RDMA_CORE_CAP_IW_CM)
560 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
561 | RDMA_CORE_CAP_OPA_MAD)
563 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
565 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
567 struct ib_port_attr {
569 enum ib_port_state state;
571 enum ib_mtu active_mtu;
573 unsigned int grh_required:1;
574 unsigned int ip_gids:1;
575 /* This is the value from PortInfo CapabilityMask, defined by IBA */
593 enum ib_device_modify_flags {
594 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
595 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
598 #define IB_DEVICE_NODE_DESC_MAX 64
600 struct ib_device_modify {
602 char node_desc[IB_DEVICE_NODE_DESC_MAX];
605 enum ib_port_modify_flags {
606 IB_PORT_SHUTDOWN = 1,
607 IB_PORT_INIT_TYPE = (1<<2),
608 IB_PORT_RESET_QKEY_CNTR = (1<<3),
609 IB_PORT_OPA_MASK_CHG = (1<<4)
612 struct ib_port_modify {
613 u32 set_port_cap_mask;
614 u32 clr_port_cap_mask;
622 IB_EVENT_QP_ACCESS_ERR,
626 IB_EVENT_PATH_MIG_ERR,
627 IB_EVENT_DEVICE_FATAL,
628 IB_EVENT_PORT_ACTIVE,
631 IB_EVENT_PKEY_CHANGE,
634 IB_EVENT_SRQ_LIMIT_REACHED,
635 IB_EVENT_QP_LAST_WQE_REACHED,
636 IB_EVENT_CLIENT_REREGISTER,
641 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
644 struct ib_device *device;
652 enum ib_event_type event;
655 struct ib_event_handler {
656 struct ib_device *device;
657 void (*handler)(struct ib_event_handler *, struct ib_event *);
658 struct list_head list;
661 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
663 (_ptr)->device = _device; \
664 (_ptr)->handler = _handler; \
665 INIT_LIST_HEAD(&(_ptr)->list); \
668 struct ib_global_route {
669 const struct ib_gid_attr *sgid_attr;
678 __be32 version_tclass_flow;
686 union rdma_network_hdr {
689 /* The IB spec states that if it's IPv4, the header
690 * is located in the last 20 bytes of the header.
693 struct iphdr roce4grh;
697 #define IB_QPN_MASK 0xFFFFFF
700 IB_MULTICAST_QPN = 0xffffff
703 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
704 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
711 IB_RATE_PORT_CURRENT = 0,
712 IB_RATE_2_5_GBPS = 2,
720 IB_RATE_120_GBPS = 10,
721 IB_RATE_14_GBPS = 11,
722 IB_RATE_56_GBPS = 12,
723 IB_RATE_112_GBPS = 13,
724 IB_RATE_168_GBPS = 14,
725 IB_RATE_25_GBPS = 15,
726 IB_RATE_100_GBPS = 16,
727 IB_RATE_200_GBPS = 17,
728 IB_RATE_300_GBPS = 18
732 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
733 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
734 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
735 * @rate: rate to convert.
737 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
740 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
741 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
742 * @rate: rate to convert.
744 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
748 * enum ib_mr_type - memory region type
749 * @IB_MR_TYPE_MEM_REG: memory region that is used for
750 * normal registration
751 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
752 * signature operations (data-integrity
754 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
755 * register any arbitrary sg lists (without
756 * the normal mr constraints - see
761 IB_MR_TYPE_SIGNATURE,
767 * IB_SIG_TYPE_NONE: Unprotected.
768 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
770 enum ib_signature_type {
776 * Signature T10-DIF block-guard types
777 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
778 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
780 enum ib_t10_dif_bg_type {
786 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
788 * @bg_type: T10-DIF block guard type (CRC|CSUM)
789 * @pi_interval: protection information interval.
790 * @bg: seed of guard computation.
791 * @app_tag: application tag of guard block
792 * @ref_tag: initial guard block reference tag.
793 * @ref_remap: Indicate wethear the reftag increments each block
794 * @app_escape: Indicate to skip block check if apptag=0xffff
795 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
796 * @apptag_check_mask: check bitmask of application tag.
798 struct ib_t10_dif_domain {
799 enum ib_t10_dif_bg_type bg_type;
807 u16 apptag_check_mask;
811 * struct ib_sig_domain - Parameters for signature domain
812 * @sig_type: specific signauture type
813 * @sig: union of all signature domain attributes that may
814 * be used to set domain layout.
816 struct ib_sig_domain {
817 enum ib_signature_type sig_type;
819 struct ib_t10_dif_domain dif;
824 * struct ib_sig_attrs - Parameters for signature handover operation
825 * @check_mask: bitmask for signature byte check (8 bytes)
826 * @mem: memory domain layout desciptor.
827 * @wire: wire domain layout desciptor.
829 struct ib_sig_attrs {
831 struct ib_sig_domain mem;
832 struct ib_sig_domain wire;
835 enum ib_sig_err_type {
842 * Signature check masks (8 bytes in total) according to the T10-PI standard:
843 * -------- -------- ------------
844 * | GUARD | APPTAG | REFTAG |
846 * -------- -------- ------------
849 IB_SIG_CHECK_GUARD = 0xc0,
850 IB_SIG_CHECK_APPTAG = 0x30,
851 IB_SIG_CHECK_REFTAG = 0x0f,
855 * struct ib_sig_err - signature error descriptor
858 enum ib_sig_err_type err_type;
865 enum ib_mr_status_check {
866 IB_MR_CHECK_SIG_STATUS = 1,
870 * struct ib_mr_status - Memory region status container
872 * @fail_status: Bitmask of MR checks status. For each
873 * failed check a corresponding status bit is set.
874 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
877 struct ib_mr_status {
879 struct ib_sig_err sig_err;
883 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
885 * @mult: multiple to convert.
887 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
889 enum rdma_ah_attr_type {
890 RDMA_AH_ATTR_TYPE_UNDEFINED,
891 RDMA_AH_ATTR_TYPE_IB,
892 RDMA_AH_ATTR_TYPE_ROCE,
893 RDMA_AH_ATTR_TYPE_OPA,
901 struct roce_ah_attr {
911 struct rdma_ah_attr {
912 struct ib_global_route grh;
917 enum rdma_ah_attr_type type;
919 struct ib_ah_attr ib;
920 struct roce_ah_attr roce;
921 struct opa_ah_attr opa;
929 IB_WC_LOC_EEC_OP_ERR,
934 IB_WC_LOC_ACCESS_ERR,
935 IB_WC_REM_INV_REQ_ERR,
936 IB_WC_REM_ACCESS_ERR,
939 IB_WC_RNR_RETRY_EXC_ERR,
940 IB_WC_LOC_RDD_VIOL_ERR,
941 IB_WC_REM_INV_RD_REQ_ERR,
944 IB_WC_INV_EEC_STATE_ERR,
946 IB_WC_RESP_TIMEOUT_ERR,
950 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
961 IB_WC_MASKED_COMP_SWAP,
962 IB_WC_MASKED_FETCH_ADD,
964 * Set value of IB_WC_RECV so consumers can test if a completion is a
965 * receive by testing (opcode & IB_WC_RECV).
968 IB_WC_RECV_RDMA_WITH_IMM
973 IB_WC_WITH_IMM = (1<<1),
974 IB_WC_WITH_INVALIDATE = (1<<2),
975 IB_WC_IP_CSUM_OK = (1<<3),
976 IB_WC_WITH_SMAC = (1<<4),
977 IB_WC_WITH_VLAN = (1<<5),
978 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
984 struct ib_cqe *wr_cqe;
986 enum ib_wc_status status;
987 enum ib_wc_opcode opcode;
1001 u8 port_num; /* valid only for DR SMPs on switches */
1004 u8 network_hdr_type;
1007 enum ib_cq_notify_flags {
1008 IB_CQ_SOLICITED = 1 << 0,
1009 IB_CQ_NEXT_COMP = 1 << 1,
1010 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1011 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1020 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1022 return srq_type == IB_SRQT_XRC ||
1023 srq_type == IB_SRQT_TM;
1026 enum ib_srq_attr_mask {
1027 IB_SRQ_MAX_WR = 1 << 0,
1028 IB_SRQ_LIMIT = 1 << 1,
1031 struct ib_srq_attr {
1037 struct ib_srq_init_attr {
1038 void (*event_handler)(struct ib_event *, void *);
1040 struct ib_srq_attr attr;
1041 enum ib_srq_type srq_type;
1047 struct ib_xrcd *xrcd;
1062 u32 max_inline_data;
1065 * Maximum number of rdma_rw_ctx structures in flight at a time.
1066 * ib_create_qp() will calculate the right amount of neededed WRs
1067 * and MRs based on this.
1079 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1080 * here (and in that order) since the MAD layer uses them as
1081 * indices into a 2-entry table.
1090 IB_QPT_RAW_ETHERTYPE,
1091 IB_QPT_RAW_PACKET = 8,
1095 IB_QPT_DRIVER = 0xFF,
1096 /* Reserve a range for qp types internal to the low level driver.
1097 * These qp types will not be visible at the IB core layer, so the
1098 * IB_QPT_MAX usages should not be affected in the core layer
1100 IB_QPT_RESERVED1 = 0x1000,
1112 enum ib_qp_create_flags {
1113 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1114 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1115 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1116 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1117 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1118 IB_QP_CREATE_NETIF_QP = 1 << 5,
1119 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1120 /* FREE = 1 << 7, */
1121 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1122 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1123 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1124 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1125 /* reserve bits 26-31 for low level drivers' internal use */
1126 IB_QP_CREATE_RESERVED_START = 1 << 26,
1127 IB_QP_CREATE_RESERVED_END = 1 << 31,
1131 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1132 * callback to destroy the passed in QP.
1135 struct ib_qp_init_attr {
1136 void (*event_handler)(struct ib_event *, void *);
1138 struct ib_cq *send_cq;
1139 struct ib_cq *recv_cq;
1141 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1142 struct ib_qp_cap cap;
1143 enum ib_sig_type sq_sig_type;
1144 enum ib_qp_type qp_type;
1145 enum ib_qp_create_flags create_flags;
1148 * Only needed for special QP types, or when using the RW API.
1151 struct ib_rwq_ind_table *rwq_ind_tbl;
1155 struct ib_qp_open_attr {
1156 void (*event_handler)(struct ib_event *, void *);
1159 enum ib_qp_type qp_type;
1162 enum ib_rnr_timeout {
1163 IB_RNR_TIMER_655_36 = 0,
1164 IB_RNR_TIMER_000_01 = 1,
1165 IB_RNR_TIMER_000_02 = 2,
1166 IB_RNR_TIMER_000_03 = 3,
1167 IB_RNR_TIMER_000_04 = 4,
1168 IB_RNR_TIMER_000_06 = 5,
1169 IB_RNR_TIMER_000_08 = 6,
1170 IB_RNR_TIMER_000_12 = 7,
1171 IB_RNR_TIMER_000_16 = 8,
1172 IB_RNR_TIMER_000_24 = 9,
1173 IB_RNR_TIMER_000_32 = 10,
1174 IB_RNR_TIMER_000_48 = 11,
1175 IB_RNR_TIMER_000_64 = 12,
1176 IB_RNR_TIMER_000_96 = 13,
1177 IB_RNR_TIMER_001_28 = 14,
1178 IB_RNR_TIMER_001_92 = 15,
1179 IB_RNR_TIMER_002_56 = 16,
1180 IB_RNR_TIMER_003_84 = 17,
1181 IB_RNR_TIMER_005_12 = 18,
1182 IB_RNR_TIMER_007_68 = 19,
1183 IB_RNR_TIMER_010_24 = 20,
1184 IB_RNR_TIMER_015_36 = 21,
1185 IB_RNR_TIMER_020_48 = 22,
1186 IB_RNR_TIMER_030_72 = 23,
1187 IB_RNR_TIMER_040_96 = 24,
1188 IB_RNR_TIMER_061_44 = 25,
1189 IB_RNR_TIMER_081_92 = 26,
1190 IB_RNR_TIMER_122_88 = 27,
1191 IB_RNR_TIMER_163_84 = 28,
1192 IB_RNR_TIMER_245_76 = 29,
1193 IB_RNR_TIMER_327_68 = 30,
1194 IB_RNR_TIMER_491_52 = 31
1197 enum ib_qp_attr_mask {
1199 IB_QP_CUR_STATE = (1<<1),
1200 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1201 IB_QP_ACCESS_FLAGS = (1<<3),
1202 IB_QP_PKEY_INDEX = (1<<4),
1203 IB_QP_PORT = (1<<5),
1204 IB_QP_QKEY = (1<<6),
1206 IB_QP_PATH_MTU = (1<<8),
1207 IB_QP_TIMEOUT = (1<<9),
1208 IB_QP_RETRY_CNT = (1<<10),
1209 IB_QP_RNR_RETRY = (1<<11),
1210 IB_QP_RQ_PSN = (1<<12),
1211 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1212 IB_QP_ALT_PATH = (1<<14),
1213 IB_QP_MIN_RNR_TIMER = (1<<15),
1214 IB_QP_SQ_PSN = (1<<16),
1215 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1216 IB_QP_PATH_MIG_STATE = (1<<18),
1217 IB_QP_CAP = (1<<19),
1218 IB_QP_DEST_QPN = (1<<20),
1219 IB_QP_RESERVED1 = (1<<21),
1220 IB_QP_RESERVED2 = (1<<22),
1221 IB_QP_RESERVED3 = (1<<23),
1222 IB_QP_RESERVED4 = (1<<24),
1223 IB_QP_RATE_LIMIT = (1<<25),
1248 enum ib_qp_state qp_state;
1249 enum ib_qp_state cur_qp_state;
1250 enum ib_mtu path_mtu;
1251 enum ib_mig_state path_mig_state;
1256 int qp_access_flags;
1257 struct ib_qp_cap cap;
1258 struct rdma_ah_attr ah_attr;
1259 struct rdma_ah_attr alt_ah_attr;
1262 u8 en_sqd_async_notify;
1265 u8 max_dest_rd_atomic;
1278 IB_WR_RDMA_WRITE_WITH_IMM,
1280 IB_WR_SEND_WITH_IMM,
1282 IB_WR_ATOMIC_CMP_AND_SWP,
1283 IB_WR_ATOMIC_FETCH_AND_ADD,
1285 IB_WR_SEND_WITH_INV,
1286 IB_WR_RDMA_READ_WITH_INV,
1289 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1290 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1292 /* reserve values for low level drivers' internal use.
1293 * These values will not be used at all in the ib core layer.
1295 IB_WR_RESERVED1 = 0xf0,
1307 enum ib_send_flags {
1309 IB_SEND_SIGNALED = (1<<1),
1310 IB_SEND_SOLICITED = (1<<2),
1311 IB_SEND_INLINE = (1<<3),
1312 IB_SEND_IP_CSUM = (1<<4),
1314 /* reserve bits 26-31 for low level drivers' internal use */
1315 IB_SEND_RESERVED_START = (1 << 26),
1316 IB_SEND_RESERVED_END = (1 << 31),
1326 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1330 struct ib_send_wr *next;
1333 struct ib_cqe *wr_cqe;
1335 struct ib_sge *sg_list;
1337 enum ib_wr_opcode opcode;
1341 u32 invalidate_rkey;
1346 struct ib_send_wr wr;
1351 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1353 return container_of(wr, struct ib_rdma_wr, wr);
1356 struct ib_atomic_wr {
1357 struct ib_send_wr wr;
1361 u64 compare_add_mask;
1366 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1368 return container_of(wr, struct ib_atomic_wr, wr);
1372 struct ib_send_wr wr;
1379 u16 pkey_index; /* valid for GSI only */
1380 u8 port_num; /* valid for DR SMPs on switch only */
1383 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1385 return container_of(wr, struct ib_ud_wr, wr);
1389 struct ib_send_wr wr;
1395 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1397 return container_of(wr, struct ib_reg_wr, wr);
1400 struct ib_sig_handover_wr {
1401 struct ib_send_wr wr;
1402 struct ib_sig_attrs *sig_attrs;
1403 struct ib_mr *sig_mr;
1405 struct ib_sge *prot;
1408 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1410 return container_of(wr, struct ib_sig_handover_wr, wr);
1414 struct ib_recv_wr *next;
1417 struct ib_cqe *wr_cqe;
1419 struct ib_sge *sg_list;
1423 enum ib_access_flags {
1424 IB_ACCESS_LOCAL_WRITE = 1,
1425 IB_ACCESS_REMOTE_WRITE = (1<<1),
1426 IB_ACCESS_REMOTE_READ = (1<<2),
1427 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1428 IB_ACCESS_MW_BIND = (1<<4),
1429 IB_ZERO_BASED = (1<<5),
1430 IB_ACCESS_ON_DEMAND = (1<<6),
1431 IB_ACCESS_HUGETLB = (1<<7),
1435 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1436 * are hidden here instead of a uapi header!
1438 enum ib_mr_rereg_flags {
1439 IB_MR_REREG_TRANS = 1,
1440 IB_MR_REREG_PD = (1<<1),
1441 IB_MR_REREG_ACCESS = (1<<2),
1442 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1445 struct ib_fmr_attr {
1453 enum rdma_remove_reason {
1455 * Userspace requested uobject deletion or initial try
1456 * to remove uobject via cleanup. Call could fail
1458 RDMA_REMOVE_DESTROY,
1459 /* Context deletion. This call should delete the actual object itself */
1461 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1462 RDMA_REMOVE_DRIVER_REMOVE,
1463 /* Context is being cleaned-up, but commit was just completed */
1464 RDMA_REMOVE_DURING_CLEANUP,
1467 struct ib_rdmacg_object {
1468 #ifdef CONFIG_CGROUP_RDMA
1469 struct rdma_cgroup *cg; /* owner rdma cgroup */
1473 struct ib_ucontext {
1474 struct ib_device *device;
1475 struct ib_uverbs_file *ufile;
1478 bool cleanup_retryable;
1481 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1482 struct rb_root_cached umem_tree;
1484 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1485 * mmu notifiers registration.
1487 struct rw_semaphore umem_rwsem;
1488 void (*invalidate_range)(struct ib_umem *umem,
1489 unsigned long start, unsigned long end);
1491 struct mmu_notifier mn;
1492 atomic_t notifier_count;
1493 /* A list of umems that don't have private mmu notifier counters yet. */
1494 struct list_head no_private_counters;
1498 struct ib_rdmacg_object cg_obj;
1502 u64 user_handle; /* handle given to us by userspace */
1503 /* ufile & ucontext owning this object */
1504 struct ib_uverbs_file *ufile;
1505 /* FIXME, save memory: ufile->context == context */
1506 struct ib_ucontext *context; /* associated user context */
1507 void *object; /* containing object */
1508 struct list_head list; /* link to context's list */
1509 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1510 int id; /* index into kernel idr */
1512 atomic_t usecnt; /* protects exclusive access */
1513 struct rcu_head rcu; /* kfree_rcu() overhead */
1515 const struct uverbs_obj_type *type;
1519 const void __user *inbuf;
1520 void __user *outbuf;
1528 struct ib_device *device;
1529 struct ib_uobject *uobject;
1530 atomic_t usecnt; /* count all resources */
1532 u32 unsafe_global_rkey;
1535 * Implementation details of the RDMA core, don't use in drivers:
1537 struct ib_mr *__internal_mr;
1538 struct rdma_restrack_entry res;
1542 struct ib_device *device;
1543 atomic_t usecnt; /* count all exposed resources */
1544 struct inode *inode;
1546 struct mutex tgt_qp_mutex;
1547 struct list_head tgt_qp_list;
1551 struct ib_device *device;
1553 struct ib_uobject *uobject;
1554 const struct ib_gid_attr *sgid_attr;
1555 enum rdma_ah_attr_type type;
1558 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1560 enum ib_poll_context {
1561 IB_POLL_DIRECT, /* caller context, no hw completions */
1562 IB_POLL_SOFTIRQ, /* poll from softirq context */
1563 IB_POLL_WORKQUEUE, /* poll from workqueue */
1567 struct ib_device *device;
1568 struct ib_uobject *uobject;
1569 ib_comp_handler comp_handler;
1570 void (*event_handler)(struct ib_event *, void *);
1573 atomic_t usecnt; /* count number of work queues */
1574 enum ib_poll_context poll_ctx;
1577 struct irq_poll iop;
1578 struct work_struct work;
1581 * Implementation details of the RDMA core, don't use in drivers:
1583 struct rdma_restrack_entry res;
1587 struct ib_device *device;
1589 struct ib_uobject *uobject;
1590 void (*event_handler)(struct ib_event *, void *);
1592 enum ib_srq_type srq_type;
1599 struct ib_xrcd *xrcd;
1606 enum ib_raw_packet_caps {
1607 /* Strip cvlan from incoming packet and report it in the matching work
1608 * completion is supported.
1610 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1611 /* Scatter FCS field of an incoming packet to host memory is supported.
1613 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1614 /* Checksum offloads are supported (for both send and receive). */
1615 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1616 /* When a packet is received for an RQ with no receive WQEs, the
1617 * packet processing is delayed.
1619 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1633 struct ib_device *device;
1634 struct ib_uobject *uobject;
1636 void (*event_handler)(struct ib_event *, void *);
1640 enum ib_wq_state state;
1641 enum ib_wq_type wq_type;
1646 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1647 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1648 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1649 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1652 struct ib_wq_init_attr {
1654 enum ib_wq_type wq_type;
1658 void (*event_handler)(struct ib_event *, void *);
1659 u32 create_flags; /* Use enum ib_wq_flags */
1662 enum ib_wq_attr_mask {
1663 IB_WQ_STATE = 1 << 0,
1664 IB_WQ_CUR_STATE = 1 << 1,
1665 IB_WQ_FLAGS = 1 << 2,
1669 enum ib_wq_state wq_state;
1670 enum ib_wq_state curr_wq_state;
1671 u32 flags; /* Use enum ib_wq_flags */
1672 u32 flags_mask; /* Use enum ib_wq_flags */
1675 struct ib_rwq_ind_table {
1676 struct ib_device *device;
1677 struct ib_uobject *uobject;
1680 u32 log_ind_tbl_size;
1681 struct ib_wq **ind_tbl;
1684 struct ib_rwq_ind_table_init_attr {
1685 u32 log_ind_tbl_size;
1686 /* Each entry is a pointer to Receive Work Queue */
1687 struct ib_wq **ind_tbl;
1690 enum port_pkey_state {
1691 IB_PORT_PKEY_NOT_VALID = 0,
1692 IB_PORT_PKEY_VALID = 1,
1693 IB_PORT_PKEY_LISTED = 2,
1696 struct ib_qp_security;
1698 struct ib_port_pkey {
1699 enum port_pkey_state state;
1702 struct list_head qp_list;
1703 struct list_head to_error_list;
1704 struct ib_qp_security *sec;
1707 struct ib_ports_pkeys {
1708 struct ib_port_pkey main;
1709 struct ib_port_pkey alt;
1712 struct ib_qp_security {
1714 struct ib_device *dev;
1715 /* Hold this mutex when changing port and pkey settings. */
1717 struct ib_ports_pkeys *ports_pkeys;
1718 /* A list of all open shared QP handles. Required to enforce security
1719 * properly for all users of a shared QP.
1721 struct list_head shared_qp_list;
1724 atomic_t error_list_count;
1725 struct completion error_complete;
1726 int error_comps_pending;
1730 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1731 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1734 struct ib_device *device;
1736 struct ib_cq *send_cq;
1737 struct ib_cq *recv_cq;
1740 struct list_head rdma_mrs;
1741 struct list_head sig_mrs;
1743 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1744 struct list_head xrcd_list;
1746 /* count times opened, mcast attaches, flow attaches */
1748 struct list_head open_list;
1749 struct ib_qp *real_qp;
1750 struct ib_uobject *uobject;
1751 void (*event_handler)(struct ib_event *, void *);
1753 /* sgid_attrs associated with the AV's */
1754 const struct ib_gid_attr *av_sgid_attr;
1755 const struct ib_gid_attr *alt_path_sgid_attr;
1759 enum ib_qp_type qp_type;
1760 struct ib_rwq_ind_table *rwq_ind_tbl;
1761 struct ib_qp_security *qp_sec;
1765 * Implementation details of the RDMA core, don't use in drivers:
1767 struct rdma_restrack_entry res;
1771 struct ib_device *device;
1774 struct ib_uobject *uobject;
1779 struct ib_device *device;
1785 unsigned int page_size;
1788 struct ib_uobject *uobject; /* user */
1789 struct list_head qp_entry; /* FR */
1795 * Implementation details of the RDMA core, don't use in drivers:
1797 struct rdma_restrack_entry res;
1801 struct ib_device *device;
1803 struct ib_uobject *uobject;
1805 enum ib_mw_type type;
1809 struct ib_device *device;
1811 struct list_head list;
1816 /* Supported steering options */
1817 enum ib_flow_attr_type {
1818 /* steering according to rule specifications */
1819 IB_FLOW_ATTR_NORMAL = 0x0,
1820 /* default unicast and multicast rule -
1821 * receive all Eth traffic which isn't steered to any QP
1823 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1824 /* default multicast rule -
1825 * receive all Eth multicast traffic which isn't steered to any QP
1827 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1828 /* sniffer rule - receive all port traffic */
1829 IB_FLOW_ATTR_SNIFFER = 0x3
1832 /* Supported steering header types */
1833 enum ib_flow_spec_type {
1835 IB_FLOW_SPEC_ETH = 0x20,
1836 IB_FLOW_SPEC_IB = 0x22,
1838 IB_FLOW_SPEC_IPV4 = 0x30,
1839 IB_FLOW_SPEC_IPV6 = 0x31,
1840 IB_FLOW_SPEC_ESP = 0x34,
1842 IB_FLOW_SPEC_TCP = 0x40,
1843 IB_FLOW_SPEC_UDP = 0x41,
1844 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1845 IB_FLOW_SPEC_GRE = 0x51,
1846 IB_FLOW_SPEC_MPLS = 0x60,
1847 IB_FLOW_SPEC_INNER = 0x100,
1849 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1850 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1851 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1852 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1854 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1855 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1857 /* Flow steering rule priority is set according to it's domain.
1858 * Lower domain value means higher priority.
1860 enum ib_flow_domain {
1861 IB_FLOW_DOMAIN_USER,
1862 IB_FLOW_DOMAIN_ETHTOOL,
1865 IB_FLOW_DOMAIN_NUM /* Must be last */
1868 enum ib_flow_flags {
1869 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1870 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1871 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1874 struct ib_flow_eth_filter {
1883 struct ib_flow_spec_eth {
1886 struct ib_flow_eth_filter val;
1887 struct ib_flow_eth_filter mask;
1890 struct ib_flow_ib_filter {
1897 struct ib_flow_spec_ib {
1900 struct ib_flow_ib_filter val;
1901 struct ib_flow_ib_filter mask;
1904 /* IPv4 header flags */
1905 enum ib_ipv4_flags {
1906 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1907 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1908 last have this flag set */
1911 struct ib_flow_ipv4_filter {
1922 struct ib_flow_spec_ipv4 {
1925 struct ib_flow_ipv4_filter val;
1926 struct ib_flow_ipv4_filter mask;
1929 struct ib_flow_ipv6_filter {
1940 struct ib_flow_spec_ipv6 {
1943 struct ib_flow_ipv6_filter val;
1944 struct ib_flow_ipv6_filter mask;
1947 struct ib_flow_tcp_udp_filter {
1954 struct ib_flow_spec_tcp_udp {
1957 struct ib_flow_tcp_udp_filter val;
1958 struct ib_flow_tcp_udp_filter mask;
1961 struct ib_flow_tunnel_filter {
1966 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1967 * the tunnel_id from val has the vni value
1969 struct ib_flow_spec_tunnel {
1972 struct ib_flow_tunnel_filter val;
1973 struct ib_flow_tunnel_filter mask;
1976 struct ib_flow_esp_filter {
1983 struct ib_flow_spec_esp {
1986 struct ib_flow_esp_filter val;
1987 struct ib_flow_esp_filter mask;
1990 struct ib_flow_gre_filter {
1991 __be16 c_ks_res0_ver;
1998 struct ib_flow_spec_gre {
2001 struct ib_flow_gre_filter val;
2002 struct ib_flow_gre_filter mask;
2005 struct ib_flow_mpls_filter {
2011 struct ib_flow_spec_mpls {
2014 struct ib_flow_mpls_filter val;
2015 struct ib_flow_mpls_filter mask;
2018 struct ib_flow_spec_action_tag {
2019 enum ib_flow_spec_type type;
2024 struct ib_flow_spec_action_drop {
2025 enum ib_flow_spec_type type;
2029 struct ib_flow_spec_action_handle {
2030 enum ib_flow_spec_type type;
2032 struct ib_flow_action *act;
2035 enum ib_counters_description {
2040 struct ib_flow_spec_action_count {
2041 enum ib_flow_spec_type type;
2043 struct ib_counters *counters;
2046 union ib_flow_spec {
2051 struct ib_flow_spec_eth eth;
2052 struct ib_flow_spec_ib ib;
2053 struct ib_flow_spec_ipv4 ipv4;
2054 struct ib_flow_spec_tcp_udp tcp_udp;
2055 struct ib_flow_spec_ipv6 ipv6;
2056 struct ib_flow_spec_tunnel tunnel;
2057 struct ib_flow_spec_esp esp;
2058 struct ib_flow_spec_gre gre;
2059 struct ib_flow_spec_mpls mpls;
2060 struct ib_flow_spec_action_tag flow_tag;
2061 struct ib_flow_spec_action_drop drop;
2062 struct ib_flow_spec_action_handle action;
2063 struct ib_flow_spec_action_count flow_count;
2066 struct ib_flow_attr {
2067 enum ib_flow_attr_type type;
2073 union ib_flow_spec flows[];
2078 struct ib_uobject *uobject;
2081 enum ib_flow_action_type {
2082 IB_FLOW_ACTION_UNSPECIFIED,
2083 IB_FLOW_ACTION_ESP = 1,
2086 struct ib_flow_action_attrs_esp_keymats {
2087 enum ib_uverbs_flow_action_esp_keymat protocol;
2089 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2093 struct ib_flow_action_attrs_esp_replays {
2094 enum ib_uverbs_flow_action_esp_replay protocol;
2096 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2100 enum ib_flow_action_attrs_esp_flags {
2101 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2102 * This is done in order to share the same flags between user-space and
2103 * kernel and spare an unnecessary translation.
2107 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2108 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2111 struct ib_flow_spec_list {
2112 struct ib_flow_spec_list *next;
2113 union ib_flow_spec spec;
2116 struct ib_flow_action_attrs_esp {
2117 struct ib_flow_action_attrs_esp_keymats *keymat;
2118 struct ib_flow_action_attrs_esp_replays *replay;
2119 struct ib_flow_spec_list *encap;
2120 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2121 * Value of 0 is a valid value.
2127 /* Use enum ib_flow_action_attrs_esp_flags */
2129 u64 hard_limit_pkts;
2132 struct ib_flow_action {
2133 struct ib_device *device;
2134 struct ib_uobject *uobject;
2135 enum ib_flow_action_type type;
2142 enum ib_process_mad_flags {
2143 IB_MAD_IGNORE_MKEY = 1,
2144 IB_MAD_IGNORE_BKEY = 2,
2145 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2148 enum ib_mad_result {
2149 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2150 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2151 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2152 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2155 struct ib_port_cache {
2157 struct ib_pkey_cache *pkey;
2158 struct ib_gid_table *gid;
2160 enum ib_port_state port_state;
2165 struct ib_event_handler event_handler;
2166 struct ib_port_cache *ports;
2171 struct ib_port_immutable {
2178 /* rdma netdev type - specifies protocol type */
2179 enum rdma_netdev_t {
2180 RDMA_NETDEV_OPA_VNIC,
2185 * struct rdma_netdev - rdma netdev
2186 * For cases where netstack interfacing is required.
2188 struct rdma_netdev {
2190 struct ib_device *hca;
2193 /* cleanup function must be specified */
2194 void (*free_rdma_netdev)(struct net_device *netdev);
2196 /* control functions */
2197 void (*set_id)(struct net_device *netdev, int id);
2199 int (*send)(struct net_device *dev, struct sk_buff *skb,
2200 struct ib_ah *address, u32 dqpn);
2202 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2203 union ib_gid *gid, u16 mlid,
2204 int set_qkey, u32 qkey);
2205 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2206 union ib_gid *gid, u16 mlid);
2209 struct ib_port_pkey_list {
2210 /* Lock to hold while modifying the list. */
2211 spinlock_t list_lock;
2212 struct list_head pkey_list;
2215 struct ib_counters {
2216 struct ib_device *device;
2217 struct ib_uobject *uobject;
2218 /* num of objects attached */
2222 enum ib_read_counters_flags {
2223 /* prefer read values from driver cache */
2224 IB_READ_COUNTERS_ATTR_PREFER_CACHED = 1 << 0,
2227 struct ib_counters_read_attr {
2230 u32 flags; /* use enum ib_read_counters_flags */
2233 struct uverbs_attr_bundle;
2236 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2237 struct device *dma_device;
2239 char name[IB_DEVICE_NAME_MAX];
2241 struct list_head event_handler_list;
2242 spinlock_t event_handler_lock;
2244 spinlock_t client_data_lock;
2245 struct list_head core_list;
2246 /* Access to the client_data_list is protected by the client_data_lock
2247 * spinlock and the lists_rwsem read-write semaphore */
2248 struct list_head client_data_list;
2250 struct ib_cache cache;
2252 * port_immutable is indexed by port number
2254 struct ib_port_immutable *port_immutable;
2256 int num_comp_vectors;
2258 struct ib_port_pkey_list *port_pkey_list;
2260 struct iw_cm_verbs *iwcm;
2263 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2264 * driver initialized data. The struct is kfree()'ed by the sysfs
2265 * core when the device is removed. A lifespan of -1 in the return
2266 * struct tells the core to set a default lifespan.
2268 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2271 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2272 * @index - The index in the value array we wish to have updated, or
2273 * num_counters if we want all stats updated
2275 * < 0 - Error, no counters updated
2276 * index - Updated the single counter pointed to by index
2277 * num_counters - Updated all counters (will reset the timestamp
2278 * and prevent further calls for lifespan milliseconds)
2279 * Drivers are allowed to update all counters in leiu of just the
2280 * one given in index at their option
2282 int (*get_hw_stats)(struct ib_device *device,
2283 struct rdma_hw_stats *stats,
2284 u8 port, int index);
2285 int (*query_device)(struct ib_device *device,
2286 struct ib_device_attr *device_attr,
2287 struct ib_udata *udata);
2288 int (*query_port)(struct ib_device *device,
2290 struct ib_port_attr *port_attr);
2291 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2293 /* When calling get_netdev, the HW vendor's driver should return the
2294 * net device of device @device at port @port_num or NULL if such
2295 * a net device doesn't exist. The vendor driver should call dev_hold
2296 * on this net device. The HW vendor's device driver must guarantee
2297 * that this function returns NULL before the net device has finished
2298 * NETDEV_UNREGISTER state.
2300 struct net_device *(*get_netdev)(struct ib_device *device,
2302 /* query_gid should be return GID value for @device, when @port_num
2303 * link layer is either IB or iWarp. It is no-op if @port_num port
2304 * is RoCE link layer.
2306 int (*query_gid)(struct ib_device *device,
2307 u8 port_num, int index,
2309 /* When calling add_gid, the HW vendor's driver should add the gid
2310 * of device of port at gid index available at @attr. Meta-info of
2311 * that gid (for example, the network device related to this gid) is
2312 * available at @attr. @context allows the HW vendor driver to store
2313 * extra information together with a GID entry. The HW vendor driver may
2314 * allocate memory to contain this information and store it in @context
2315 * when a new GID entry is written to. Params are consistent until the
2316 * next call of add_gid or delete_gid. The function should return 0 on
2317 * success or error otherwise. The function could be called
2318 * concurrently for different ports. This function is only called when
2319 * roce_gid_table is used.
2321 int (*add_gid)(const struct ib_gid_attr *attr,
2323 /* When calling del_gid, the HW vendor's driver should delete the
2324 * gid of device @device at gid index gid_index of port port_num
2325 * available in @attr.
2326 * Upon the deletion of a GID entry, the HW vendor must free any
2327 * allocated memory. The caller will clear @context afterwards.
2328 * This function is only called when roce_gid_table is used.
2330 int (*del_gid)(const struct ib_gid_attr *attr,
2332 int (*query_pkey)(struct ib_device *device,
2333 u8 port_num, u16 index, u16 *pkey);
2334 int (*modify_device)(struct ib_device *device,
2335 int device_modify_mask,
2336 struct ib_device_modify *device_modify);
2337 int (*modify_port)(struct ib_device *device,
2338 u8 port_num, int port_modify_mask,
2339 struct ib_port_modify *port_modify);
2340 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2341 struct ib_udata *udata);
2342 int (*dealloc_ucontext)(struct ib_ucontext *context);
2343 int (*mmap)(struct ib_ucontext *context,
2344 struct vm_area_struct *vma);
2345 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2346 struct ib_ucontext *context,
2347 struct ib_udata *udata);
2348 int (*dealloc_pd)(struct ib_pd *pd);
2349 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2350 struct rdma_ah_attr *ah_attr,
2351 struct ib_udata *udata);
2352 int (*modify_ah)(struct ib_ah *ah,
2353 struct rdma_ah_attr *ah_attr);
2354 int (*query_ah)(struct ib_ah *ah,
2355 struct rdma_ah_attr *ah_attr);
2356 int (*destroy_ah)(struct ib_ah *ah);
2357 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2358 struct ib_srq_init_attr *srq_init_attr,
2359 struct ib_udata *udata);
2360 int (*modify_srq)(struct ib_srq *srq,
2361 struct ib_srq_attr *srq_attr,
2362 enum ib_srq_attr_mask srq_attr_mask,
2363 struct ib_udata *udata);
2364 int (*query_srq)(struct ib_srq *srq,
2365 struct ib_srq_attr *srq_attr);
2366 int (*destroy_srq)(struct ib_srq *srq);
2367 int (*post_srq_recv)(struct ib_srq *srq,
2368 struct ib_recv_wr *recv_wr,
2369 struct ib_recv_wr **bad_recv_wr);
2370 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2371 struct ib_qp_init_attr *qp_init_attr,
2372 struct ib_udata *udata);
2373 int (*modify_qp)(struct ib_qp *qp,
2374 struct ib_qp_attr *qp_attr,
2376 struct ib_udata *udata);
2377 int (*query_qp)(struct ib_qp *qp,
2378 struct ib_qp_attr *qp_attr,
2380 struct ib_qp_init_attr *qp_init_attr);
2381 int (*destroy_qp)(struct ib_qp *qp);
2382 int (*post_send)(struct ib_qp *qp,
2383 struct ib_send_wr *send_wr,
2384 struct ib_send_wr **bad_send_wr);
2385 int (*post_recv)(struct ib_qp *qp,
2386 struct ib_recv_wr *recv_wr,
2387 struct ib_recv_wr **bad_recv_wr);
2388 struct ib_cq * (*create_cq)(struct ib_device *device,
2389 const struct ib_cq_init_attr *attr,
2390 struct ib_ucontext *context,
2391 struct ib_udata *udata);
2392 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2394 int (*destroy_cq)(struct ib_cq *cq);
2395 int (*resize_cq)(struct ib_cq *cq, int cqe,
2396 struct ib_udata *udata);
2397 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2399 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2400 int (*req_notify_cq)(struct ib_cq *cq,
2401 enum ib_cq_notify_flags flags);
2402 int (*req_ncomp_notif)(struct ib_cq *cq,
2404 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2405 int mr_access_flags);
2406 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2407 u64 start, u64 length,
2409 int mr_access_flags,
2410 struct ib_udata *udata);
2411 int (*rereg_user_mr)(struct ib_mr *mr,
2413 u64 start, u64 length,
2415 int mr_access_flags,
2417 struct ib_udata *udata);
2418 int (*dereg_mr)(struct ib_mr *mr);
2419 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2420 enum ib_mr_type mr_type,
2422 int (*map_mr_sg)(struct ib_mr *mr,
2423 struct scatterlist *sg,
2425 unsigned int *sg_offset);
2426 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2427 enum ib_mw_type type,
2428 struct ib_udata *udata);
2429 int (*dealloc_mw)(struct ib_mw *mw);
2430 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2431 int mr_access_flags,
2432 struct ib_fmr_attr *fmr_attr);
2433 int (*map_phys_fmr)(struct ib_fmr *fmr,
2434 u64 *page_list, int list_len,
2436 int (*unmap_fmr)(struct list_head *fmr_list);
2437 int (*dealloc_fmr)(struct ib_fmr *fmr);
2438 int (*attach_mcast)(struct ib_qp *qp,
2441 int (*detach_mcast)(struct ib_qp *qp,
2444 int (*process_mad)(struct ib_device *device,
2445 int process_mad_flags,
2447 const struct ib_wc *in_wc,
2448 const struct ib_grh *in_grh,
2449 const struct ib_mad_hdr *in_mad,
2451 struct ib_mad_hdr *out_mad,
2452 size_t *out_mad_size,
2453 u16 *out_mad_pkey_index);
2454 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2455 struct ib_ucontext *ucontext,
2456 struct ib_udata *udata);
2457 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2458 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2462 struct ib_udata *udata);
2463 int (*destroy_flow)(struct ib_flow *flow_id);
2464 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2465 struct ib_mr_status *mr_status);
2466 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2467 void (*drain_rq)(struct ib_qp *qp);
2468 void (*drain_sq)(struct ib_qp *qp);
2469 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2471 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2472 struct ifla_vf_info *ivf);
2473 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2474 struct ifla_vf_stats *stats);
2475 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2477 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2478 struct ib_wq_init_attr *init_attr,
2479 struct ib_udata *udata);
2480 int (*destroy_wq)(struct ib_wq *wq);
2481 int (*modify_wq)(struct ib_wq *wq,
2482 struct ib_wq_attr *attr,
2484 struct ib_udata *udata);
2485 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2486 struct ib_rwq_ind_table_init_attr *init_attr,
2487 struct ib_udata *udata);
2488 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2489 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2490 const struct ib_flow_action_attrs_esp *attr,
2491 struct uverbs_attr_bundle *attrs);
2492 int (*destroy_flow_action)(struct ib_flow_action *action);
2493 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2494 const struct ib_flow_action_attrs_esp *attr,
2495 struct uverbs_attr_bundle *attrs);
2496 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2497 struct ib_ucontext *context,
2498 struct ib_dm_alloc_attr *attr,
2499 struct uverbs_attr_bundle *attrs);
2500 int (*dealloc_dm)(struct ib_dm *dm);
2501 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2502 struct ib_dm_mr_attr *attr,
2503 struct uverbs_attr_bundle *attrs);
2504 struct ib_counters * (*create_counters)(struct ib_device *device,
2505 struct uverbs_attr_bundle *attrs);
2506 int (*destroy_counters)(struct ib_counters *counters);
2507 int (*read_counters)(struct ib_counters *counters,
2508 struct ib_counters_read_attr *counters_read_attr,
2509 struct uverbs_attr_bundle *attrs);
2512 * rdma netdev operation
2514 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2515 * doesn't support the specified rdma netdev type.
2517 struct net_device *(*alloc_rdma_netdev)(
2518 struct ib_device *device,
2520 enum rdma_netdev_t type,
2522 unsigned char name_assign_type,
2523 void (*setup)(struct net_device *));
2525 struct module *owner;
2527 struct kobject *ports_parent;
2528 struct list_head port_list;
2531 IB_DEV_UNINITIALIZED,
2537 u64 uverbs_cmd_mask;
2538 u64 uverbs_ex_cmd_mask;
2540 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2546 struct ib_device_attr attrs;
2547 struct attribute_group *hw_stats_ag;
2548 struct rdma_hw_stats *hw_stats;
2550 #ifdef CONFIG_CGROUP_RDMA
2551 struct rdmacg_device cg_device;
2556 * Implementation details of the RDMA core, don't use in drivers
2558 struct rdma_restrack_root res;
2561 * The following mandatory functions are used only at device
2562 * registration. Keep functions such as these at the end of this
2563 * structure to avoid cache line misses when accessing struct ib_device
2566 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2567 void (*get_dev_fw_str)(struct ib_device *, char *str);
2568 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2571 struct uverbs_root_spec *driver_specs_root;
2572 enum rdma_driver_id driver_id;
2577 void (*add) (struct ib_device *);
2578 void (*remove)(struct ib_device *, void *client_data);
2580 /* Returns the net_dev belonging to this ib_client and matching the
2582 * @dev: An RDMA device that the net_dev use for communication.
2583 * @port: A physical port number on the RDMA device.
2584 * @pkey: P_Key that the net_dev uses if applicable.
2585 * @gid: A GID that the net_dev uses to communicate.
2586 * @addr: An IP address the net_dev is configured with.
2587 * @client_data: The device's client data set by ib_set_client_data().
2589 * An ib_client that implements a net_dev on top of RDMA devices
2590 * (such as IP over IB) should implement this callback, allowing the
2591 * rdma_cm module to find the right net_dev for a given request.
2593 * The caller is responsible for calling dev_put on the returned
2595 struct net_device *(*get_net_dev_by_params)(
2596 struct ib_device *dev,
2599 const union ib_gid *gid,
2600 const struct sockaddr *addr,
2602 struct list_head list;
2605 struct ib_device *ib_alloc_device(size_t size);
2606 void ib_dealloc_device(struct ib_device *device);
2608 void ib_get_device_fw_str(struct ib_device *device, char *str);
2610 int ib_register_device(struct ib_device *device,
2611 int (*port_callback)(struct ib_device *,
2612 u8, struct kobject *));
2613 void ib_unregister_device(struct ib_device *device);
2615 int ib_register_client (struct ib_client *client);
2616 void ib_unregister_client(struct ib_client *client);
2618 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2619 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2622 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2624 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2627 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2629 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2632 static inline bool ib_is_buffer_cleared(const void __user *p,
2638 if (len > USHRT_MAX)
2641 buf = memdup_user(p, len);
2645 ret = !memchr_inv(buf, 0, len);
2650 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2654 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2658 * ib_is_destroy_retryable - Check whether the uobject destruction
2660 * @ret: The initial destruction return code
2661 * @why: remove reason
2662 * @uobj: The uobject that is destroyed
2664 * This function is a helper function that IB layer and low-level drivers
2665 * can use to consider whether the destruction of the given uobject is
2667 * It checks the original return code, if it wasn't success the destruction
2668 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2669 * the remove reason. (i.e. why).
2670 * Must be called with the object locked for destroy.
2672 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2673 struct ib_uobject *uobj)
2675 return ret && (why == RDMA_REMOVE_DESTROY ||
2676 uobj->context->cleanup_retryable);
2680 * ib_destroy_usecnt - Called during destruction to check the usecnt
2681 * @usecnt: The usecnt atomic
2682 * @why: remove reason
2683 * @uobj: The uobject that is destroyed
2685 * Non-zero usecnts will block destruction unless destruction was triggered by
2686 * a ucontext cleanup.
2688 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2689 enum rdma_remove_reason why,
2690 struct ib_uobject *uobj)
2692 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2698 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2699 * contains all required attributes and no attributes not allowed for
2700 * the given QP state transition.
2701 * @cur_state: Current QP state
2702 * @next_state: Next QP state
2704 * @mask: Mask of supplied QP attributes
2705 * @ll : link layer of port
2707 * This function is a helper function that a low-level driver's
2708 * modify_qp method can use to validate the consumer's input. It
2709 * checks that cur_state and next_state are valid QP states, that a
2710 * transition from cur_state to next_state is allowed by the IB spec,
2711 * and that the attribute mask supplied is allowed for the transition.
2713 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2714 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2715 enum rdma_link_layer ll);
2717 void ib_register_event_handler(struct ib_event_handler *event_handler);
2718 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2719 void ib_dispatch_event(struct ib_event *event);
2721 int ib_query_port(struct ib_device *device,
2722 u8 port_num, struct ib_port_attr *port_attr);
2724 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2728 * rdma_cap_ib_switch - Check if the device is IB switch
2729 * @device: Device to check
2731 * Device driver is responsible for setting is_switch bit on
2732 * in ib_device structure at init time.
2734 * Return: true if the device is IB switch.
2736 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2738 return device->is_switch;
2742 * rdma_start_port - Return the first valid port number for the device
2745 * @device: Device to be checked
2747 * Return start port number
2749 static inline u8 rdma_start_port(const struct ib_device *device)
2751 return rdma_cap_ib_switch(device) ? 0 : 1;
2755 * rdma_end_port - Return the last valid port number for the device
2758 * @device: Device to be checked
2760 * Return last port number
2762 static inline u8 rdma_end_port(const struct ib_device *device)
2764 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2767 static inline int rdma_is_port_valid(const struct ib_device *device,
2770 return (port >= rdma_start_port(device) &&
2771 port <= rdma_end_port(device));
2774 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2776 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2779 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2781 return device->port_immutable[port_num].core_cap_flags &
2782 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2785 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2787 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2790 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2792 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2795 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2797 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2800 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2802 return rdma_protocol_ib(device, port_num) ||
2803 rdma_protocol_roce(device, port_num);
2806 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2808 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2811 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2813 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2817 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2818 * Management Datagrams.
2819 * @device: Device to check
2820 * @port_num: Port number to check
2822 * Management Datagrams (MAD) are a required part of the InfiniBand
2823 * specification and are supported on all InfiniBand devices. A slightly
2824 * extended version are also supported on OPA interfaces.
2826 * Return: true if the port supports sending/receiving of MAD packets.
2828 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2830 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2834 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2835 * Management Datagrams.
2836 * @device: Device to check
2837 * @port_num: Port number to check
2839 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2840 * datagrams with their own versions. These OPA MADs share many but not all of
2841 * the characteristics of InfiniBand MADs.
2843 * OPA MADs differ in the following ways:
2845 * 1) MADs are variable size up to 2K
2846 * IBTA defined MADs remain fixed at 256 bytes
2847 * 2) OPA SMPs must carry valid PKeys
2848 * 3) OPA SMP packets are a different format
2850 * Return: true if the port supports OPA MAD packet formats.
2852 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2854 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2855 == RDMA_CORE_CAP_OPA_MAD;
2859 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2860 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2861 * @device: Device to check
2862 * @port_num: Port number to check
2864 * Each InfiniBand node is required to provide a Subnet Management Agent
2865 * that the subnet manager can access. Prior to the fabric being fully
2866 * configured by the subnet manager, the SMA is accessed via a well known
2867 * interface called the Subnet Management Interface (SMI). This interface
2868 * uses directed route packets to communicate with the SM to get around the
2869 * chicken and egg problem of the SM needing to know what's on the fabric
2870 * in order to configure the fabric, and needing to configure the fabric in
2871 * order to send packets to the devices on the fabric. These directed
2872 * route packets do not need the fabric fully configured in order to reach
2873 * their destination. The SMI is the only method allowed to send
2874 * directed route packets on an InfiniBand fabric.
2876 * Return: true if the port provides an SMI.
2878 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2880 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2884 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2885 * Communication Manager.
2886 * @device: Device to check
2887 * @port_num: Port number to check
2889 * The InfiniBand Communication Manager is one of many pre-defined General
2890 * Service Agents (GSA) that are accessed via the General Service
2891 * Interface (GSI). It's role is to facilitate establishment of connections
2892 * between nodes as well as other management related tasks for established
2895 * Return: true if the port supports an IB CM (this does not guarantee that
2896 * a CM is actually running however).
2898 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2900 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2904 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2905 * Communication Manager.
2906 * @device: Device to check
2907 * @port_num: Port number to check
2909 * Similar to above, but specific to iWARP connections which have a different
2910 * managment protocol than InfiniBand.
2912 * Return: true if the port supports an iWARP CM (this does not guarantee that
2913 * a CM is actually running however).
2915 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2917 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2921 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2922 * Subnet Administration.
2923 * @device: Device to check
2924 * @port_num: Port number to check
2926 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2927 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2928 * fabrics, devices should resolve routes to other hosts by contacting the
2929 * SA to query the proper route.
2931 * Return: true if the port should act as a client to the fabric Subnet
2932 * Administration interface. This does not imply that the SA service is
2935 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2937 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2941 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2943 * @device: Device to check
2944 * @port_num: Port number to check
2946 * InfiniBand multicast registration is more complex than normal IPv4 or
2947 * IPv6 multicast registration. Each Host Channel Adapter must register
2948 * with the Subnet Manager when it wishes to join a multicast group. It
2949 * should do so only once regardless of how many queue pairs it subscribes
2950 * to this group. And it should leave the group only after all queue pairs
2951 * attached to the group have been detached.
2953 * Return: true if the port must undertake the additional adminstrative
2954 * overhead of registering/unregistering with the SM and tracking of the
2955 * total number of queue pairs attached to the multicast group.
2957 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2959 return rdma_cap_ib_sa(device, port_num);
2963 * rdma_cap_af_ib - Check if the port of device has the capability
2964 * Native Infiniband Address.
2965 * @device: Device to check
2966 * @port_num: Port number to check
2968 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2969 * GID. RoCE uses a different mechanism, but still generates a GID via
2970 * a prescribed mechanism and port specific data.
2972 * Return: true if the port uses a GID address to identify devices on the
2975 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2977 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2981 * rdma_cap_eth_ah - Check if the port of device has the capability
2982 * Ethernet Address Handle.
2983 * @device: Device to check
2984 * @port_num: Port number to check
2986 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2987 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2988 * port. Normally, packet headers are generated by the sending host
2989 * adapter, but when sending connectionless datagrams, we must manually
2990 * inject the proper headers for the fabric we are communicating over.
2992 * Return: true if we are running as a RoCE port and must force the
2993 * addition of a Global Route Header built from our Ethernet Address
2994 * Handle into our header list for connectionless packets.
2996 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2998 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3002 * rdma_cap_opa_ah - Check if the port of device supports
3003 * OPA Address handles
3004 * @device: Device to check
3005 * @port_num: Port number to check
3007 * Return: true if we are running on an OPA device which supports
3008 * the extended OPA addressing.
3010 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3012 return (device->port_immutable[port_num].core_cap_flags &
3013 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3017 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3020 * @port_num: Port number
3022 * This MAD size includes the MAD headers and MAD payload. No other headers
3025 * Return the max MAD size required by the Port. Will return 0 if the port
3026 * does not support MADs
3028 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3030 return device->port_immutable[port_num].max_mad_size;
3034 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3035 * @device: Device to check
3036 * @port_num: Port number to check
3038 * RoCE GID table mechanism manages the various GIDs for a device.
3040 * NOTE: if allocating the port's GID table has failed, this call will still
3041 * return true, but any RoCE GID table API will fail.
3043 * Return: true if the port uses RoCE GID table mechanism in order to manage
3046 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3049 return rdma_protocol_roce(device, port_num) &&
3050 device->add_gid && device->del_gid;
3054 * Check if the device supports READ W/ INVALIDATE.
3056 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3059 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3060 * has support for it yet.
3062 return rdma_protocol_iwarp(dev, port_num);
3065 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3067 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3068 struct ifla_vf_info *info);
3069 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3070 struct ifla_vf_stats *stats);
3071 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3074 int ib_query_pkey(struct ib_device *device,
3075 u8 port_num, u16 index, u16 *pkey);
3077 int ib_modify_device(struct ib_device *device,
3078 int device_modify_mask,
3079 struct ib_device_modify *device_modify);
3081 int ib_modify_port(struct ib_device *device,
3082 u8 port_num, int port_modify_mask,
3083 struct ib_port_modify *port_modify);
3085 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3086 u8 *port_num, u16 *index);
3088 int ib_find_pkey(struct ib_device *device,
3089 u8 port_num, u16 pkey, u16 *index);
3093 * Create a memory registration for all memory in the system and place
3094 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3095 * ULPs to avoid the overhead of dynamic MRs.
3097 * This flag is generally considered unsafe and must only be used in
3098 * extremly trusted environments. Every use of it will log a warning
3099 * in the kernel log.
3101 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3104 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3105 const char *caller);
3106 #define ib_alloc_pd(device, flags) \
3107 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3108 void ib_dealloc_pd(struct ib_pd *pd);
3111 * rdma_create_ah - Creates an address handle for the given address vector.
3112 * @pd: The protection domain associated with the address handle.
3113 * @ah_attr: The attributes of the address vector.
3115 * The address handle is used to reference a local or global destination
3116 * in all UD QP post sends.
3118 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3121 * rdma_create_user_ah - Creates an address handle for the given address vector.
3122 * It resolves destination mac address for ah attribute of RoCE type.
3123 * @pd: The protection domain associated with the address handle.
3124 * @ah_attr: The attributes of the address vector.
3125 * @udata: pointer to user's input output buffer information need by
3128 * It returns 0 on success and returns appropriate error code on error.
3129 * The address handle is used to reference a local or global destination
3130 * in all UD QP post sends.
3132 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3133 struct rdma_ah_attr *ah_attr,
3134 struct ib_udata *udata);
3136 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3138 * @hdr: the L3 header to parse
3139 * @net_type: type of header to parse
3140 * @sgid: place to store source gid
3141 * @dgid: place to store destination gid
3143 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3144 enum rdma_network_type net_type,
3145 union ib_gid *sgid, union ib_gid *dgid);
3148 * ib_get_rdma_header_version - Get the header version
3149 * @hdr: the L3 header to parse
3151 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3154 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3156 * @device: Device on which the received message arrived.
3157 * @port_num: Port on which the received message arrived.
3158 * @wc: Work completion associated with the received message.
3159 * @grh: References the received global route header. This parameter is
3160 * ignored unless the work completion indicates that the GRH is valid.
3161 * @ah_attr: Returned attributes that can be used when creating an address
3162 * handle for replying to the message.
3163 * When ib_init_ah_attr_from_wc() returns success,
3164 * (a) for IB link layer it optionally contains a reference to SGID attribute
3165 * when GRH is present for IB link layer.
3166 * (b) for RoCE link layer it contains a reference to SGID attribute.
3167 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3168 * attributes which are initialized using ib_init_ah_attr_from_wc().
3171 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3172 const struct ib_wc *wc, const struct ib_grh *grh,
3173 struct rdma_ah_attr *ah_attr);
3176 * ib_create_ah_from_wc - Creates an address handle associated with the
3177 * sender of the specified work completion.
3178 * @pd: The protection domain associated with the address handle.
3179 * @wc: Work completion information associated with a received message.
3180 * @grh: References the received global route header. This parameter is
3181 * ignored unless the work completion indicates that the GRH is valid.
3182 * @port_num: The outbound port number to associate with the address.
3184 * The address handle is used to reference a local or global destination
3185 * in all UD QP post sends.
3187 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3188 const struct ib_grh *grh, u8 port_num);
3191 * rdma_modify_ah - Modifies the address vector associated with an address
3193 * @ah: The address handle to modify.
3194 * @ah_attr: The new address vector attributes to associate with the
3197 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3200 * rdma_query_ah - Queries the address vector associated with an address
3202 * @ah: The address handle to query.
3203 * @ah_attr: The address vector attributes associated with the address
3206 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3209 * rdma_destroy_ah - Destroys an address handle.
3210 * @ah: The address handle to destroy.
3212 int rdma_destroy_ah(struct ib_ah *ah);
3215 * ib_create_srq - Creates a SRQ associated with the specified protection
3217 * @pd: The protection domain associated with the SRQ.
3218 * @srq_init_attr: A list of initial attributes required to create the
3219 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3220 * the actual capabilities of the created SRQ.
3222 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3223 * requested size of the SRQ, and set to the actual values allocated
3224 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3225 * will always be at least as large as the requested values.
3227 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3228 struct ib_srq_init_attr *srq_init_attr);
3231 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3232 * @srq: The SRQ to modify.
3233 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3234 * the current values of selected SRQ attributes are returned.
3235 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3236 * are being modified.
3238 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3239 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3240 * the number of receives queued drops below the limit.
3242 int ib_modify_srq(struct ib_srq *srq,
3243 struct ib_srq_attr *srq_attr,
3244 enum ib_srq_attr_mask srq_attr_mask);
3247 * ib_query_srq - Returns the attribute list and current values for the
3249 * @srq: The SRQ to query.
3250 * @srq_attr: The attributes of the specified SRQ.
3252 int ib_query_srq(struct ib_srq *srq,
3253 struct ib_srq_attr *srq_attr);
3256 * ib_destroy_srq - Destroys the specified SRQ.
3257 * @srq: The SRQ to destroy.
3259 int ib_destroy_srq(struct ib_srq *srq);
3262 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3263 * @srq: The SRQ to post the work request on.
3264 * @recv_wr: A list of work requests to post on the receive queue.
3265 * @bad_recv_wr: On an immediate failure, this parameter will reference
3266 * the work request that failed to be posted on the QP.
3268 static inline int ib_post_srq_recv(struct ib_srq *srq,
3269 struct ib_recv_wr *recv_wr,
3270 struct ib_recv_wr **bad_recv_wr)
3272 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3276 * ib_create_qp - Creates a QP associated with the specified protection
3278 * @pd: The protection domain associated with the QP.
3279 * @qp_init_attr: A list of initial attributes required to create the
3280 * QP. If QP creation succeeds, then the attributes are updated to
3281 * the actual capabilities of the created QP.
3283 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3284 struct ib_qp_init_attr *qp_init_attr);
3287 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3288 * @qp: The QP to modify.
3289 * @attr: On input, specifies the QP attributes to modify. On output,
3290 * the current values of selected QP attributes are returned.
3291 * @attr_mask: A bit-mask used to specify which attributes of the QP
3292 * are being modified.
3293 * @udata: pointer to user's input output buffer information
3294 * are being modified.
3295 * It returns 0 on success and returns appropriate error code on error.
3297 int ib_modify_qp_with_udata(struct ib_qp *qp,
3298 struct ib_qp_attr *attr,
3300 struct ib_udata *udata);
3303 * ib_modify_qp - Modifies the attributes for the specified QP and then
3304 * transitions the QP to the given state.
3305 * @qp: The QP to modify.
3306 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3307 * the current values of selected QP attributes are returned.
3308 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3309 * are being modified.
3311 int ib_modify_qp(struct ib_qp *qp,
3312 struct ib_qp_attr *qp_attr,
3316 * ib_query_qp - Returns the attribute list and current values for the
3318 * @qp: The QP to query.
3319 * @qp_attr: The attributes of the specified QP.
3320 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3321 * @qp_init_attr: Additional attributes of the selected QP.
3323 * The qp_attr_mask may be used to limit the query to gathering only the
3324 * selected attributes.
3326 int ib_query_qp(struct ib_qp *qp,
3327 struct ib_qp_attr *qp_attr,
3329 struct ib_qp_init_attr *qp_init_attr);
3332 * ib_destroy_qp - Destroys the specified QP.
3333 * @qp: The QP to destroy.
3335 int ib_destroy_qp(struct ib_qp *qp);
3338 * ib_open_qp - Obtain a reference to an existing sharable QP.
3339 * @xrcd - XRC domain
3340 * @qp_open_attr: Attributes identifying the QP to open.
3342 * Returns a reference to a sharable QP.
3344 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3345 struct ib_qp_open_attr *qp_open_attr);
3348 * ib_close_qp - Release an external reference to a QP.
3349 * @qp: The QP handle to release
3351 * The opened QP handle is released by the caller. The underlying
3352 * shared QP is not destroyed until all internal references are released.
3354 int ib_close_qp(struct ib_qp *qp);
3357 * ib_post_send - Posts a list of work requests to the send queue of
3359 * @qp: The QP to post the work request on.
3360 * @send_wr: A list of work requests to post on the send queue.
3361 * @bad_send_wr: On an immediate failure, this parameter will reference
3362 * the work request that failed to be posted on the QP.
3364 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3365 * error is returned, the QP state shall not be affected,
3366 * ib_post_send() will return an immediate error after queueing any
3367 * earlier work requests in the list.
3369 static inline int ib_post_send(struct ib_qp *qp,
3370 struct ib_send_wr *send_wr,
3371 struct ib_send_wr **bad_send_wr)
3373 return qp->device->post_send(qp, send_wr, bad_send_wr);
3377 * ib_post_recv - Posts a list of work requests to the receive queue of
3379 * @qp: The QP to post the work request on.
3380 * @recv_wr: A list of work requests to post on the receive queue.
3381 * @bad_recv_wr: On an immediate failure, this parameter will reference
3382 * the work request that failed to be posted on the QP.
3384 static inline int ib_post_recv(struct ib_qp *qp,
3385 struct ib_recv_wr *recv_wr,
3386 struct ib_recv_wr **bad_recv_wr)
3388 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3391 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3392 int nr_cqe, int comp_vector,
3393 enum ib_poll_context poll_ctx, const char *caller);
3394 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3395 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3397 void ib_free_cq(struct ib_cq *cq);
3398 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3401 * ib_create_cq - Creates a CQ on the specified device.
3402 * @device: The device on which to create the CQ.
3403 * @comp_handler: A user-specified callback that is invoked when a
3404 * completion event occurs on the CQ.
3405 * @event_handler: A user-specified callback that is invoked when an
3406 * asynchronous event not associated with a completion occurs on the CQ.
3407 * @cq_context: Context associated with the CQ returned to the user via
3408 * the associated completion and event handlers.
3409 * @cq_attr: The attributes the CQ should be created upon.
3411 * Users can examine the cq structure to determine the actual CQ size.
3413 struct ib_cq *ib_create_cq(struct ib_device *device,
3414 ib_comp_handler comp_handler,
3415 void (*event_handler)(struct ib_event *, void *),
3417 const struct ib_cq_init_attr *cq_attr);
3420 * ib_resize_cq - Modifies the capacity of the CQ.
3421 * @cq: The CQ to resize.
3422 * @cqe: The minimum size of the CQ.
3424 * Users can examine the cq structure to determine the actual CQ size.
3426 int ib_resize_cq(struct ib_cq *cq, int cqe);
3429 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3430 * @cq: The CQ to modify.
3431 * @cq_count: number of CQEs that will trigger an event
3432 * @cq_period: max period of time in usec before triggering an event
3435 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3438 * ib_destroy_cq - Destroys the specified CQ.
3439 * @cq: The CQ to destroy.
3441 int ib_destroy_cq(struct ib_cq *cq);
3444 * ib_poll_cq - poll a CQ for completion(s)
3445 * @cq:the CQ being polled
3446 * @num_entries:maximum number of completions to return
3447 * @wc:array of at least @num_entries &struct ib_wc where completions
3450 * Poll a CQ for (possibly multiple) completions. If the return value
3451 * is < 0, an error occurred. If the return value is >= 0, it is the
3452 * number of completions returned. If the return value is
3453 * non-negative and < num_entries, then the CQ was emptied.
3455 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3458 return cq->device->poll_cq(cq, num_entries, wc);
3462 * ib_req_notify_cq - Request completion notification on a CQ.
3463 * @cq: The CQ to generate an event for.
3465 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3466 * to request an event on the next solicited event or next work
3467 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3468 * may also be |ed in to request a hint about missed events, as
3472 * < 0 means an error occurred while requesting notification
3473 * == 0 means notification was requested successfully, and if
3474 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3475 * were missed and it is safe to wait for another event. In
3476 * this case is it guaranteed that any work completions added
3477 * to the CQ since the last CQ poll will trigger a completion
3478 * notification event.
3479 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3480 * in. It means that the consumer must poll the CQ again to
3481 * make sure it is empty to avoid missing an event because of a
3482 * race between requesting notification and an entry being
3483 * added to the CQ. This return value means it is possible
3484 * (but not guaranteed) that a work completion has been added
3485 * to the CQ since the last poll without triggering a
3486 * completion notification event.
3488 static inline int ib_req_notify_cq(struct ib_cq *cq,
3489 enum ib_cq_notify_flags flags)
3491 return cq->device->req_notify_cq(cq, flags);
3495 * ib_req_ncomp_notif - Request completion notification when there are
3496 * at least the specified number of unreaped completions on the CQ.
3497 * @cq: The CQ to generate an event for.
3498 * @wc_cnt: The number of unreaped completions that should be on the
3499 * CQ before an event is generated.
3501 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3503 return cq->device->req_ncomp_notif ?
3504 cq->device->req_ncomp_notif(cq, wc_cnt) :
3509 * ib_dma_mapping_error - check a DMA addr for error
3510 * @dev: The device for which the dma_addr was created
3511 * @dma_addr: The DMA address to check
3513 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3515 return dma_mapping_error(dev->dma_device, dma_addr);
3519 * ib_dma_map_single - Map a kernel virtual address to DMA address
3520 * @dev: The device for which the dma_addr is to be created
3521 * @cpu_addr: The kernel virtual address
3522 * @size: The size of the region in bytes
3523 * @direction: The direction of the DMA
3525 static inline u64 ib_dma_map_single(struct ib_device *dev,
3526 void *cpu_addr, size_t size,
3527 enum dma_data_direction direction)
3529 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3533 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3534 * @dev: The device for which the DMA address was created
3535 * @addr: The DMA address
3536 * @size: The size of the region in bytes
3537 * @direction: The direction of the DMA
3539 static inline void ib_dma_unmap_single(struct ib_device *dev,
3540 u64 addr, size_t size,
3541 enum dma_data_direction direction)
3543 dma_unmap_single(dev->dma_device, addr, size, direction);
3547 * ib_dma_map_page - Map a physical page to DMA address
3548 * @dev: The device for which the dma_addr is to be created
3549 * @page: The page to be mapped
3550 * @offset: The offset within the page
3551 * @size: The size of the region in bytes
3552 * @direction: The direction of the DMA
3554 static inline u64 ib_dma_map_page(struct ib_device *dev,
3556 unsigned long offset,
3558 enum dma_data_direction direction)
3560 return dma_map_page(dev->dma_device, page, offset, size, direction);
3564 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3565 * @dev: The device for which the DMA address was created
3566 * @addr: The DMA address
3567 * @size: The size of the region in bytes
3568 * @direction: The direction of the DMA
3570 static inline void ib_dma_unmap_page(struct ib_device *dev,
3571 u64 addr, size_t size,
3572 enum dma_data_direction direction)
3574 dma_unmap_page(dev->dma_device, addr, size, direction);
3578 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3579 * @dev: The device for which the DMA addresses are to be created
3580 * @sg: The array of scatter/gather entries
3581 * @nents: The number of scatter/gather entries
3582 * @direction: The direction of the DMA
3584 static inline int ib_dma_map_sg(struct ib_device *dev,
3585 struct scatterlist *sg, int nents,
3586 enum dma_data_direction direction)
3588 return dma_map_sg(dev->dma_device, sg, nents, direction);
3592 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3593 * @dev: The device for which the DMA addresses were created
3594 * @sg: The array of scatter/gather entries
3595 * @nents: The number of scatter/gather entries
3596 * @direction: The direction of the DMA
3598 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3599 struct scatterlist *sg, int nents,
3600 enum dma_data_direction direction)
3602 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3605 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3606 struct scatterlist *sg, int nents,
3607 enum dma_data_direction direction,
3608 unsigned long dma_attrs)
3610 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3614 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3615 struct scatterlist *sg, int nents,
3616 enum dma_data_direction direction,
3617 unsigned long dma_attrs)
3619 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3622 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3623 * @dev: The device for which the DMA addresses were created
3624 * @sg: The scatter/gather entry
3626 * Note: this function is obsolete. To do: change all occurrences of
3627 * ib_sg_dma_address() into sg_dma_address().
3629 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3630 struct scatterlist *sg)
3632 return sg_dma_address(sg);
3636 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3637 * @dev: The device for which the DMA addresses were created
3638 * @sg: The scatter/gather entry
3640 * Note: this function is obsolete. To do: change all occurrences of
3641 * ib_sg_dma_len() into sg_dma_len().
3643 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3644 struct scatterlist *sg)
3646 return sg_dma_len(sg);
3650 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3651 * @dev: The device for which the DMA address was created
3652 * @addr: The DMA address
3653 * @size: The size of the region in bytes
3654 * @dir: The direction of the DMA
3656 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3659 enum dma_data_direction dir)
3661 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3665 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3666 * @dev: The device for which the DMA address was created
3667 * @addr: The DMA address
3668 * @size: The size of the region in bytes
3669 * @dir: The direction of the DMA
3671 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3674 enum dma_data_direction dir)
3676 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3680 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3681 * @dev: The device for which the DMA address is requested
3682 * @size: The size of the region to allocate in bytes
3683 * @dma_handle: A pointer for returning the DMA address of the region
3684 * @flag: memory allocator flags
3686 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3688 dma_addr_t *dma_handle,
3691 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3695 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3696 * @dev: The device for which the DMA addresses were allocated
3697 * @size: The size of the region
3698 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3699 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3701 static inline void ib_dma_free_coherent(struct ib_device *dev,
3702 size_t size, void *cpu_addr,
3703 dma_addr_t dma_handle)
3705 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3709 * ib_dereg_mr - Deregisters a memory region and removes it from the
3710 * HCA translation table.
3711 * @mr: The memory region to deregister.
3713 * This function can fail, if the memory region has memory windows bound to it.
3715 int ib_dereg_mr(struct ib_mr *mr);
3717 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3718 enum ib_mr_type mr_type,
3722 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3724 * @mr - struct ib_mr pointer to be updated.
3725 * @newkey - new key to be used.
3727 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3729 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3730 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3734 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3735 * for calculating a new rkey for type 2 memory windows.
3736 * @rkey - the rkey to increment.
3738 static inline u32 ib_inc_rkey(u32 rkey)
3740 const u32 mask = 0x000000ff;
3741 return ((rkey + 1) & mask) | (rkey & ~mask);
3745 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3746 * @pd: The protection domain associated with the unmapped region.
3747 * @mr_access_flags: Specifies the memory access rights.
3748 * @fmr_attr: Attributes of the unmapped region.
3750 * A fast memory region must be mapped before it can be used as part of
3753 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3754 int mr_access_flags,
3755 struct ib_fmr_attr *fmr_attr);
3758 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3759 * @fmr: The fast memory region to associate with the pages.
3760 * @page_list: An array of physical pages to map to the fast memory region.
3761 * @list_len: The number of pages in page_list.
3762 * @iova: The I/O virtual address to use with the mapped region.
3764 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3765 u64 *page_list, int list_len,
3768 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3772 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3773 * @fmr_list: A linked list of fast memory regions to unmap.
3775 int ib_unmap_fmr(struct list_head *fmr_list);
3778 * ib_dealloc_fmr - Deallocates a fast memory region.
3779 * @fmr: The fast memory region to deallocate.
3781 int ib_dealloc_fmr(struct ib_fmr *fmr);
3784 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3785 * @qp: QP to attach to the multicast group. The QP must be type
3787 * @gid: Multicast group GID.
3788 * @lid: Multicast group LID in host byte order.
3790 * In order to send and receive multicast packets, subnet
3791 * administration must have created the multicast group and configured
3792 * the fabric appropriately. The port associated with the specified
3793 * QP must also be a member of the multicast group.
3795 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3798 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3799 * @qp: QP to detach from the multicast group.
3800 * @gid: Multicast group GID.
3801 * @lid: Multicast group LID in host byte order.
3803 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3806 * ib_alloc_xrcd - Allocates an XRC domain.
3807 * @device: The device on which to allocate the XRC domain.
3808 * @caller: Module name for kernel consumers
3810 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3811 #define ib_alloc_xrcd(device) \
3812 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3815 * ib_dealloc_xrcd - Deallocates an XRC domain.
3816 * @xrcd: The XRC domain to deallocate.
3818 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3820 static inline int ib_check_mr_access(int flags)
3823 * Local write permission is required if remote write or
3824 * remote atomic permission is also requested.
3826 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3827 !(flags & IB_ACCESS_LOCAL_WRITE))
3833 static inline bool ib_access_writable(int access_flags)
3836 * We have writable memory backing the MR if any of the following
3837 * access flags are set. "Local write" and "remote write" obviously
3838 * require write access. "Remote atomic" can do things like fetch and
3839 * add, which will modify memory, and "MW bind" can change permissions
3840 * by binding a window.
3842 return access_flags &
3843 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3844 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3848 * ib_check_mr_status: lightweight check of MR status.
3849 * This routine may provide status checks on a selected
3850 * ib_mr. first use is for signature status check.
3852 * @mr: A memory region.
3853 * @check_mask: Bitmask of which checks to perform from
3854 * ib_mr_status_check enumeration.
3855 * @mr_status: The container of relevant status checks.
3856 * failed checks will be indicated in the status bitmask
3857 * and the relevant info shall be in the error item.
3859 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3860 struct ib_mr_status *mr_status);
3862 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3863 u16 pkey, const union ib_gid *gid,
3864 const struct sockaddr *addr);
3865 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3866 struct ib_wq_init_attr *init_attr);
3867 int ib_destroy_wq(struct ib_wq *wq);
3868 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3870 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3871 struct ib_rwq_ind_table_init_attr*
3872 wq_ind_table_init_attr);
3873 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3875 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3876 unsigned int *sg_offset, unsigned int page_size);
3879 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3880 unsigned int *sg_offset, unsigned int page_size)
3884 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3890 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3891 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3893 void ib_drain_rq(struct ib_qp *qp);
3894 void ib_drain_sq(struct ib_qp *qp);
3895 void ib_drain_qp(struct ib_qp *qp);
3897 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3899 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3901 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3902 return attr->roce.dmac;
3906 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3908 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3909 attr->ib.dlid = (u16)dlid;
3910 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3911 attr->opa.dlid = dlid;
3914 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3916 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3917 return attr->ib.dlid;
3918 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3919 return attr->opa.dlid;
3923 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3928 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3933 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3936 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3937 attr->ib.src_path_bits = src_path_bits;
3938 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3939 attr->opa.src_path_bits = src_path_bits;
3942 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3944 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3945 return attr->ib.src_path_bits;
3946 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3947 return attr->opa.src_path_bits;
3951 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3954 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3955 attr->opa.make_grd = make_grd;
3958 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3960 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3961 return attr->opa.make_grd;
3965 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3967 attr->port_num = port_num;
3970 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3972 return attr->port_num;
3975 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3978 attr->static_rate = static_rate;
3981 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3983 return attr->static_rate;
3986 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3987 enum ib_ah_flags flag)
3989 attr->ah_flags = flag;
3992 static inline enum ib_ah_flags
3993 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3995 return attr->ah_flags;
3998 static inline const struct ib_global_route
3999 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4004 /*To retrieve and modify the grh */
4005 static inline struct ib_global_route
4006 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4011 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4013 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4015 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4018 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4021 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4023 grh->dgid.global.subnet_prefix = prefix;
4026 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4029 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4031 grh->dgid.global.interface_id = if_id;
4034 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4035 union ib_gid *dgid, u32 flow_label,
4036 u8 sgid_index, u8 hop_limit,
4039 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4041 attr->ah_flags = IB_AH_GRH;
4044 grh->flow_label = flow_label;
4045 grh->sgid_index = sgid_index;
4046 grh->hop_limit = hop_limit;
4047 grh->traffic_class = traffic_class;
4048 grh->sgid_attr = NULL;
4051 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4052 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4053 u32 flow_label, u8 hop_limit, u8 traffic_class,
4054 const struct ib_gid_attr *sgid_attr);
4055 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4056 const struct rdma_ah_attr *src);
4057 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4058 const struct rdma_ah_attr *new);
4059 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4062 * rdma_ah_find_type - Return address handle type.
4064 * @dev: Device to be checked
4065 * @port_num: Port number
4067 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4070 if (rdma_protocol_roce(dev, port_num))
4071 return RDMA_AH_ATTR_TYPE_ROCE;
4072 if (rdma_protocol_ib(dev, port_num)) {
4073 if (rdma_cap_opa_ah(dev, port_num))
4074 return RDMA_AH_ATTR_TYPE_OPA;
4075 return RDMA_AH_ATTR_TYPE_IB;
4078 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4082 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4083 * In the current implementation the only way to get
4084 * get the 32bit lid is from other sources for OPA.
4085 * For IB, lids will always be 16bits so cast the
4086 * value accordingly.
4090 static inline u16 ib_lid_cpu16(u32 lid)
4092 WARN_ON_ONCE(lid & 0xFFFF0000);
4097 * ib_lid_be16 - Return lid in 16bit BE encoding.
4101 static inline __be16 ib_lid_be16(u32 lid)
4103 WARN_ON_ONCE(lid & 0xFFFF0000);
4104 return cpu_to_be16((u16)lid);
4108 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4110 * @device: the rdma device
4111 * @comp_vector: index of completion vector
4113 * Returns NULL on failure, otherwise a corresponding cpu map of the
4114 * completion vector (returns all-cpus map if the device driver doesn't
4115 * implement get_vector_affinity).
4117 static inline const struct cpumask *
4118 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4120 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4121 !device->get_vector_affinity)
4124 return device->get_vector_affinity(device, comp_vector);
4129 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4130 * and add their gids, as needed, to the relevant RoCE devices.
4132 * @device: the rdma device
4134 void rdma_roce_rescan_device(struct ib_device *ibdev);
4136 struct ib_ucontext *ib_uverbs_get_ucontext(struct ib_uverbs_file *ufile);
4138 #endif /* IB_VERBS_H */