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;
97 enum ib_gid_type gid_type;
102 enum rdma_node_type {
103 /* IB values map to NodeInfo:NodeType. */
113 /* set the local administered indication */
114 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
117 enum rdma_transport_type {
119 RDMA_TRANSPORT_IWARP,
120 RDMA_TRANSPORT_USNIC,
121 RDMA_TRANSPORT_USNIC_UDP
124 enum rdma_protocol_type {
128 RDMA_PROTOCOL_USNIC_UDP
131 __attribute_const__ enum rdma_transport_type
132 rdma_node_get_transport(enum rdma_node_type node_type);
134 enum rdma_network_type {
136 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
141 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
143 if (network_type == RDMA_NETWORK_IPV4 ||
144 network_type == RDMA_NETWORK_IPV6)
145 return IB_GID_TYPE_ROCE_UDP_ENCAP;
147 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
148 return IB_GID_TYPE_IB;
151 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
154 if (gid_type == IB_GID_TYPE_IB)
155 return RDMA_NETWORK_IB;
157 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
158 return RDMA_NETWORK_IPV4;
160 return RDMA_NETWORK_IPV6;
163 enum rdma_link_layer {
164 IB_LINK_LAYER_UNSPECIFIED,
165 IB_LINK_LAYER_INFINIBAND,
166 IB_LINK_LAYER_ETHERNET,
169 enum ib_device_cap_flags {
170 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
171 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
172 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
173 IB_DEVICE_RAW_MULTI = (1 << 3),
174 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
175 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
176 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
177 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
178 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
179 /* Not in use, former INIT_TYPE = (1 << 9),*/
180 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
181 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
182 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
183 IB_DEVICE_SRQ_RESIZE = (1 << 13),
184 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
187 * This device supports a per-device lkey or stag that can be
188 * used without performing a memory registration for the local
189 * memory. Note that ULPs should never check this flag, but
190 * instead of use the local_dma_lkey flag in the ib_pd structure,
191 * which will always contain a usable lkey.
193 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
194 /* Reserved, old SEND_W_INV = (1 << 16),*/
195 IB_DEVICE_MEM_WINDOW = (1 << 17),
197 * Devices should set IB_DEVICE_UD_IP_SUM if they support
198 * insertion of UDP and TCP checksum on outgoing UD IPoIB
199 * messages and can verify the validity of checksum for
200 * incoming messages. Setting this flag implies that the
201 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
203 IB_DEVICE_UD_IP_CSUM = (1 << 18),
204 IB_DEVICE_UD_TSO = (1 << 19),
205 IB_DEVICE_XRC = (1 << 20),
208 * This device supports the IB "base memory management extension",
209 * which includes support for fast registrations (IB_WR_REG_MR,
210 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
211 * also be set by any iWarp device which must support FRs to comply
212 * to the iWarp verbs spec. iWarp devices also support the
213 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
216 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
217 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
218 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
219 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
220 IB_DEVICE_RC_IP_CSUM = (1 << 25),
221 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
222 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
224 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
225 * support execution of WQEs that involve synchronization
226 * of I/O operations with single completion queue managed
229 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
230 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
231 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
232 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
233 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
234 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
235 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
236 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
237 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
238 /* The device supports padding incoming writes to cacheline. */
239 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
242 enum ib_signature_prot_cap {
243 IB_PROT_T10DIF_TYPE_1 = 1,
244 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
245 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
248 enum ib_signature_guard_cap {
249 IB_GUARD_T10DIF_CRC = 1,
250 IB_GUARD_T10DIF_CSUM = 1 << 1,
259 enum ib_odp_general_cap_bits {
260 IB_ODP_SUPPORT = 1 << 0,
261 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
264 enum ib_odp_transport_cap_bits {
265 IB_ODP_SUPPORT_SEND = 1 << 0,
266 IB_ODP_SUPPORT_RECV = 1 << 1,
267 IB_ODP_SUPPORT_WRITE = 1 << 2,
268 IB_ODP_SUPPORT_READ = 1 << 3,
269 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
273 uint64_t general_caps;
275 uint32_t rc_odp_caps;
276 uint32_t uc_odp_caps;
277 uint32_t ud_odp_caps;
278 } per_transport_caps;
282 /* Corresponding bit will be set if qp type from
283 * 'enum ib_qp_type' is supported, e.g.
284 * supported_qpts |= 1 << IB_QPT_UD
287 u32 max_rwq_indirection_tables;
288 u32 max_rwq_indirection_table_size;
291 enum ib_tm_cap_flags {
292 /* Support tag matching on RC transport */
293 IB_TM_CAP_RC = 1 << 0,
297 /* Max size of RNDV header */
298 u32 max_rndv_hdr_size;
299 /* Max number of entries in tag matching list */
301 /* From enum ib_tm_cap_flags */
303 /* Max number of outstanding list operations */
305 /* Max number of SGE in tag matching entry */
309 struct ib_cq_init_attr {
315 enum ib_cq_attr_mask {
316 IB_CQ_MODERATE = 1 << 0,
320 u16 max_cq_moderation_count;
321 u16 max_cq_moderation_period;
324 struct ib_dm_mr_attr {
330 struct ib_dm_alloc_attr {
336 struct ib_device_attr {
338 __be64 sys_image_guid;
346 u64 device_cap_flags;
356 int max_qp_init_rd_atom;
357 int max_ee_init_rd_atom;
358 enum ib_atomic_cap atomic_cap;
359 enum ib_atomic_cap masked_atomic_cap;
366 int max_mcast_qp_attach;
367 int max_total_mcast_qp_attach;
374 unsigned int max_fast_reg_page_list_len;
376 u8 local_ca_ack_delay;
379 struct ib_odp_caps odp_caps;
380 uint64_t timestamp_mask;
381 uint64_t hca_core_clock; /* in KHZ */
382 struct ib_rss_caps rss_caps;
384 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
385 struct ib_tm_caps tm_caps;
386 struct ib_cq_caps cq_caps;
398 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
401 case IB_MTU_256: return 256;
402 case IB_MTU_512: return 512;
403 case IB_MTU_1024: return 1024;
404 case IB_MTU_2048: return 2048;
405 case IB_MTU_4096: return 4096;
410 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
414 else if (mtu >= 2048)
416 else if (mtu >= 1024)
430 IB_PORT_ACTIVE_DEFER = 5
433 enum ib_port_cap_flags {
435 IB_PORT_NOTICE_SUP = 1 << 2,
436 IB_PORT_TRAP_SUP = 1 << 3,
437 IB_PORT_OPT_IPD_SUP = 1 << 4,
438 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
439 IB_PORT_SL_MAP_SUP = 1 << 6,
440 IB_PORT_MKEY_NVRAM = 1 << 7,
441 IB_PORT_PKEY_NVRAM = 1 << 8,
442 IB_PORT_LED_INFO_SUP = 1 << 9,
443 IB_PORT_SM_DISABLED = 1 << 10,
444 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
445 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
446 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
447 IB_PORT_CM_SUP = 1 << 16,
448 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
449 IB_PORT_REINIT_SUP = 1 << 18,
450 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
451 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
452 IB_PORT_DR_NOTICE_SUP = 1 << 21,
453 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
454 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
455 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
456 IB_PORT_CLIENT_REG_SUP = 1 << 25,
457 IB_PORT_IP_BASED_GIDS = 1 << 26,
467 static inline int ib_width_enum_to_int(enum ib_port_width width)
470 case IB_WIDTH_1X: return 1;
471 case IB_WIDTH_4X: return 4;
472 case IB_WIDTH_8X: return 8;
473 case IB_WIDTH_12X: return 12;
489 * struct rdma_hw_stats
490 * @lock - Mutex to protect parallel write access to lifespan and values
491 * of counters, which are 64bits and not guaranteeed to be written
492 * atomicaly on 32bits systems.
493 * @timestamp - Used by the core code to track when the last update was
494 * @lifespan - Used by the core code to determine how old the counters
495 * should be before being updated again. Stored in jiffies, defaults
496 * to 10 milliseconds, drivers can override the default be specifying
497 * their own value during their allocation routine.
498 * @name - Array of pointers to static names used for the counters in
500 * @num_counters - How many hardware counters there are. If name is
501 * shorter than this number, a kernel oops will result. Driver authors
502 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
503 * in their code to prevent this.
504 * @value - Array of u64 counters that are accessed by the sysfs code and
505 * filled in by the drivers get_stats routine
507 struct rdma_hw_stats {
508 struct mutex lock; /* Protect lifespan and values[] */
509 unsigned long timestamp;
510 unsigned long lifespan;
511 const char * const *names;
516 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
518 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
520 * @names - Array of static const char *
521 * @num_counters - How many elements in array
522 * @lifespan - How many milliseconds between updates
524 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
525 const char * const *names, int num_counters,
526 unsigned long lifespan)
528 struct rdma_hw_stats *stats;
530 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
534 stats->names = names;
535 stats->num_counters = num_counters;
536 stats->lifespan = msecs_to_jiffies(lifespan);
542 /* Define bits for the various functionality this port needs to be supported by
545 /* Management 0x00000FFF */
546 #define RDMA_CORE_CAP_IB_MAD 0x00000001
547 #define RDMA_CORE_CAP_IB_SMI 0x00000002
548 #define RDMA_CORE_CAP_IB_CM 0x00000004
549 #define RDMA_CORE_CAP_IW_CM 0x00000008
550 #define RDMA_CORE_CAP_IB_SA 0x00000010
551 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
553 /* Address format 0x000FF000 */
554 #define RDMA_CORE_CAP_AF_IB 0x00001000
555 #define RDMA_CORE_CAP_ETH_AH 0x00002000
556 #define RDMA_CORE_CAP_OPA_AH 0x00004000
558 /* Protocol 0xFFF00000 */
559 #define RDMA_CORE_CAP_PROT_IB 0x00100000
560 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
561 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
562 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
563 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
564 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
566 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
567 | RDMA_CORE_CAP_IB_MAD \
568 | RDMA_CORE_CAP_IB_SMI \
569 | RDMA_CORE_CAP_IB_CM \
570 | RDMA_CORE_CAP_IB_SA \
571 | RDMA_CORE_CAP_AF_IB)
572 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
573 | RDMA_CORE_CAP_IB_MAD \
574 | RDMA_CORE_CAP_IB_CM \
575 | RDMA_CORE_CAP_AF_IB \
576 | RDMA_CORE_CAP_ETH_AH)
577 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
578 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
579 | RDMA_CORE_CAP_IB_MAD \
580 | RDMA_CORE_CAP_IB_CM \
581 | RDMA_CORE_CAP_AF_IB \
582 | RDMA_CORE_CAP_ETH_AH)
583 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
584 | RDMA_CORE_CAP_IW_CM)
585 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
586 | RDMA_CORE_CAP_OPA_MAD)
588 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
590 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
592 struct ib_port_attr {
594 enum ib_port_state state;
596 enum ib_mtu active_mtu;
616 enum ib_device_modify_flags {
617 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
618 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
621 #define IB_DEVICE_NODE_DESC_MAX 64
623 struct ib_device_modify {
625 char node_desc[IB_DEVICE_NODE_DESC_MAX];
628 enum ib_port_modify_flags {
629 IB_PORT_SHUTDOWN = 1,
630 IB_PORT_INIT_TYPE = (1<<2),
631 IB_PORT_RESET_QKEY_CNTR = (1<<3),
632 IB_PORT_OPA_MASK_CHG = (1<<4)
635 struct ib_port_modify {
636 u32 set_port_cap_mask;
637 u32 clr_port_cap_mask;
645 IB_EVENT_QP_ACCESS_ERR,
649 IB_EVENT_PATH_MIG_ERR,
650 IB_EVENT_DEVICE_FATAL,
651 IB_EVENT_PORT_ACTIVE,
654 IB_EVENT_PKEY_CHANGE,
657 IB_EVENT_SRQ_LIMIT_REACHED,
658 IB_EVENT_QP_LAST_WQE_REACHED,
659 IB_EVENT_CLIENT_REREGISTER,
664 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
667 struct ib_device *device;
675 enum ib_event_type event;
678 struct ib_event_handler {
679 struct ib_device *device;
680 void (*handler)(struct ib_event_handler *, struct ib_event *);
681 struct list_head list;
684 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
686 (_ptr)->device = _device; \
687 (_ptr)->handler = _handler; \
688 INIT_LIST_HEAD(&(_ptr)->list); \
691 struct ib_global_route {
700 __be32 version_tclass_flow;
708 union rdma_network_hdr {
711 /* The IB spec states that if it's IPv4, the header
712 * is located in the last 20 bytes of the header.
715 struct iphdr roce4grh;
719 #define IB_QPN_MASK 0xFFFFFF
722 IB_MULTICAST_QPN = 0xffffff
725 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
726 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
733 IB_RATE_PORT_CURRENT = 0,
734 IB_RATE_2_5_GBPS = 2,
742 IB_RATE_120_GBPS = 10,
743 IB_RATE_14_GBPS = 11,
744 IB_RATE_56_GBPS = 12,
745 IB_RATE_112_GBPS = 13,
746 IB_RATE_168_GBPS = 14,
747 IB_RATE_25_GBPS = 15,
748 IB_RATE_100_GBPS = 16,
749 IB_RATE_200_GBPS = 17,
750 IB_RATE_300_GBPS = 18
754 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
755 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
756 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
757 * @rate: rate to convert.
759 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
762 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
763 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
764 * @rate: rate to convert.
766 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
770 * enum ib_mr_type - memory region type
771 * @IB_MR_TYPE_MEM_REG: memory region that is used for
772 * normal registration
773 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
774 * signature operations (data-integrity
776 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
777 * register any arbitrary sg lists (without
778 * the normal mr constraints - see
783 IB_MR_TYPE_SIGNATURE,
789 * IB_SIG_TYPE_NONE: Unprotected.
790 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
792 enum ib_signature_type {
798 * Signature T10-DIF block-guard types
799 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
800 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
802 enum ib_t10_dif_bg_type {
808 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
810 * @bg_type: T10-DIF block guard type (CRC|CSUM)
811 * @pi_interval: protection information interval.
812 * @bg: seed of guard computation.
813 * @app_tag: application tag of guard block
814 * @ref_tag: initial guard block reference tag.
815 * @ref_remap: Indicate wethear the reftag increments each block
816 * @app_escape: Indicate to skip block check if apptag=0xffff
817 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
818 * @apptag_check_mask: check bitmask of application tag.
820 struct ib_t10_dif_domain {
821 enum ib_t10_dif_bg_type bg_type;
829 u16 apptag_check_mask;
833 * struct ib_sig_domain - Parameters for signature domain
834 * @sig_type: specific signauture type
835 * @sig: union of all signature domain attributes that may
836 * be used to set domain layout.
838 struct ib_sig_domain {
839 enum ib_signature_type sig_type;
841 struct ib_t10_dif_domain dif;
846 * struct ib_sig_attrs - Parameters for signature handover operation
847 * @check_mask: bitmask for signature byte check (8 bytes)
848 * @mem: memory domain layout desciptor.
849 * @wire: wire domain layout desciptor.
851 struct ib_sig_attrs {
853 struct ib_sig_domain mem;
854 struct ib_sig_domain wire;
857 enum ib_sig_err_type {
864 * Signature check masks (8 bytes in total) according to the T10-PI standard:
865 * -------- -------- ------------
866 * | GUARD | APPTAG | REFTAG |
868 * -------- -------- ------------
871 IB_SIG_CHECK_GUARD = 0xc0,
872 IB_SIG_CHECK_APPTAG = 0x30,
873 IB_SIG_CHECK_REFTAG = 0x0f,
877 * struct ib_sig_err - signature error descriptor
880 enum ib_sig_err_type err_type;
887 enum ib_mr_status_check {
888 IB_MR_CHECK_SIG_STATUS = 1,
892 * struct ib_mr_status - Memory region status container
894 * @fail_status: Bitmask of MR checks status. For each
895 * failed check a corresponding status bit is set.
896 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
899 struct ib_mr_status {
901 struct ib_sig_err sig_err;
905 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
907 * @mult: multiple to convert.
909 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
911 enum rdma_ah_attr_type {
912 RDMA_AH_ATTR_TYPE_UNDEFINED,
913 RDMA_AH_ATTR_TYPE_IB,
914 RDMA_AH_ATTR_TYPE_ROCE,
915 RDMA_AH_ATTR_TYPE_OPA,
923 struct roce_ah_attr {
933 struct rdma_ah_attr {
934 struct ib_global_route grh;
939 enum rdma_ah_attr_type type;
941 struct ib_ah_attr ib;
942 struct roce_ah_attr roce;
943 struct opa_ah_attr opa;
951 IB_WC_LOC_EEC_OP_ERR,
956 IB_WC_LOC_ACCESS_ERR,
957 IB_WC_REM_INV_REQ_ERR,
958 IB_WC_REM_ACCESS_ERR,
961 IB_WC_RNR_RETRY_EXC_ERR,
962 IB_WC_LOC_RDD_VIOL_ERR,
963 IB_WC_REM_INV_RD_REQ_ERR,
966 IB_WC_INV_EEC_STATE_ERR,
968 IB_WC_RESP_TIMEOUT_ERR,
972 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
983 IB_WC_MASKED_COMP_SWAP,
984 IB_WC_MASKED_FETCH_ADD,
986 * Set value of IB_WC_RECV so consumers can test if a completion is a
987 * receive by testing (opcode & IB_WC_RECV).
990 IB_WC_RECV_RDMA_WITH_IMM
995 IB_WC_WITH_IMM = (1<<1),
996 IB_WC_WITH_INVALIDATE = (1<<2),
997 IB_WC_IP_CSUM_OK = (1<<3),
998 IB_WC_WITH_SMAC = (1<<4),
999 IB_WC_WITH_VLAN = (1<<5),
1000 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1006 struct ib_cqe *wr_cqe;
1008 enum ib_wc_status status;
1009 enum ib_wc_opcode opcode;
1015 u32 invalidate_rkey;
1023 u8 port_num; /* valid only for DR SMPs on switches */
1026 u8 network_hdr_type;
1029 enum ib_cq_notify_flags {
1030 IB_CQ_SOLICITED = 1 << 0,
1031 IB_CQ_NEXT_COMP = 1 << 1,
1032 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1033 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1042 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1044 return srq_type == IB_SRQT_XRC ||
1045 srq_type == IB_SRQT_TM;
1048 enum ib_srq_attr_mask {
1049 IB_SRQ_MAX_WR = 1 << 0,
1050 IB_SRQ_LIMIT = 1 << 1,
1053 struct ib_srq_attr {
1059 struct ib_srq_init_attr {
1060 void (*event_handler)(struct ib_event *, void *);
1062 struct ib_srq_attr attr;
1063 enum ib_srq_type srq_type;
1069 struct ib_xrcd *xrcd;
1084 u32 max_inline_data;
1087 * Maximum number of rdma_rw_ctx structures in flight at a time.
1088 * ib_create_qp() will calculate the right amount of neededed WRs
1089 * and MRs based on this.
1101 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1102 * here (and in that order) since the MAD layer uses them as
1103 * indices into a 2-entry table.
1112 IB_QPT_RAW_ETHERTYPE,
1113 IB_QPT_RAW_PACKET = 8,
1117 IB_QPT_DRIVER = 0xFF,
1118 /* Reserve a range for qp types internal to the low level driver.
1119 * These qp types will not be visible at the IB core layer, so the
1120 * IB_QPT_MAX usages should not be affected in the core layer
1122 IB_QPT_RESERVED1 = 0x1000,
1134 enum ib_qp_create_flags {
1135 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1136 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1137 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1138 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1139 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1140 IB_QP_CREATE_NETIF_QP = 1 << 5,
1141 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1142 /* FREE = 1 << 7, */
1143 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1144 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1145 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1146 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1147 /* reserve bits 26-31 for low level drivers' internal use */
1148 IB_QP_CREATE_RESERVED_START = 1 << 26,
1149 IB_QP_CREATE_RESERVED_END = 1 << 31,
1153 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1154 * callback to destroy the passed in QP.
1157 struct ib_qp_init_attr {
1158 void (*event_handler)(struct ib_event *, void *);
1160 struct ib_cq *send_cq;
1161 struct ib_cq *recv_cq;
1163 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1164 struct ib_qp_cap cap;
1165 enum ib_sig_type sq_sig_type;
1166 enum ib_qp_type qp_type;
1167 enum ib_qp_create_flags create_flags;
1170 * Only needed for special QP types, or when using the RW API.
1173 struct ib_rwq_ind_table *rwq_ind_tbl;
1177 struct ib_qp_open_attr {
1178 void (*event_handler)(struct ib_event *, void *);
1181 enum ib_qp_type qp_type;
1184 enum ib_rnr_timeout {
1185 IB_RNR_TIMER_655_36 = 0,
1186 IB_RNR_TIMER_000_01 = 1,
1187 IB_RNR_TIMER_000_02 = 2,
1188 IB_RNR_TIMER_000_03 = 3,
1189 IB_RNR_TIMER_000_04 = 4,
1190 IB_RNR_TIMER_000_06 = 5,
1191 IB_RNR_TIMER_000_08 = 6,
1192 IB_RNR_TIMER_000_12 = 7,
1193 IB_RNR_TIMER_000_16 = 8,
1194 IB_RNR_TIMER_000_24 = 9,
1195 IB_RNR_TIMER_000_32 = 10,
1196 IB_RNR_TIMER_000_48 = 11,
1197 IB_RNR_TIMER_000_64 = 12,
1198 IB_RNR_TIMER_000_96 = 13,
1199 IB_RNR_TIMER_001_28 = 14,
1200 IB_RNR_TIMER_001_92 = 15,
1201 IB_RNR_TIMER_002_56 = 16,
1202 IB_RNR_TIMER_003_84 = 17,
1203 IB_RNR_TIMER_005_12 = 18,
1204 IB_RNR_TIMER_007_68 = 19,
1205 IB_RNR_TIMER_010_24 = 20,
1206 IB_RNR_TIMER_015_36 = 21,
1207 IB_RNR_TIMER_020_48 = 22,
1208 IB_RNR_TIMER_030_72 = 23,
1209 IB_RNR_TIMER_040_96 = 24,
1210 IB_RNR_TIMER_061_44 = 25,
1211 IB_RNR_TIMER_081_92 = 26,
1212 IB_RNR_TIMER_122_88 = 27,
1213 IB_RNR_TIMER_163_84 = 28,
1214 IB_RNR_TIMER_245_76 = 29,
1215 IB_RNR_TIMER_327_68 = 30,
1216 IB_RNR_TIMER_491_52 = 31
1219 enum ib_qp_attr_mask {
1221 IB_QP_CUR_STATE = (1<<1),
1222 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1223 IB_QP_ACCESS_FLAGS = (1<<3),
1224 IB_QP_PKEY_INDEX = (1<<4),
1225 IB_QP_PORT = (1<<5),
1226 IB_QP_QKEY = (1<<6),
1228 IB_QP_PATH_MTU = (1<<8),
1229 IB_QP_TIMEOUT = (1<<9),
1230 IB_QP_RETRY_CNT = (1<<10),
1231 IB_QP_RNR_RETRY = (1<<11),
1232 IB_QP_RQ_PSN = (1<<12),
1233 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1234 IB_QP_ALT_PATH = (1<<14),
1235 IB_QP_MIN_RNR_TIMER = (1<<15),
1236 IB_QP_SQ_PSN = (1<<16),
1237 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1238 IB_QP_PATH_MIG_STATE = (1<<18),
1239 IB_QP_CAP = (1<<19),
1240 IB_QP_DEST_QPN = (1<<20),
1241 IB_QP_RESERVED1 = (1<<21),
1242 IB_QP_RESERVED2 = (1<<22),
1243 IB_QP_RESERVED3 = (1<<23),
1244 IB_QP_RESERVED4 = (1<<24),
1245 IB_QP_RATE_LIMIT = (1<<25),
1270 enum ib_qp_state qp_state;
1271 enum ib_qp_state cur_qp_state;
1272 enum ib_mtu path_mtu;
1273 enum ib_mig_state path_mig_state;
1278 int qp_access_flags;
1279 struct ib_qp_cap cap;
1280 struct rdma_ah_attr ah_attr;
1281 struct rdma_ah_attr alt_ah_attr;
1284 u8 en_sqd_async_notify;
1287 u8 max_dest_rd_atomic;
1300 IB_WR_RDMA_WRITE_WITH_IMM,
1302 IB_WR_SEND_WITH_IMM,
1304 IB_WR_ATOMIC_CMP_AND_SWP,
1305 IB_WR_ATOMIC_FETCH_AND_ADD,
1307 IB_WR_SEND_WITH_INV,
1308 IB_WR_RDMA_READ_WITH_INV,
1311 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1312 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1314 /* reserve values for low level drivers' internal use.
1315 * These values will not be used at all in the ib core layer.
1317 IB_WR_RESERVED1 = 0xf0,
1329 enum ib_send_flags {
1331 IB_SEND_SIGNALED = (1<<1),
1332 IB_SEND_SOLICITED = (1<<2),
1333 IB_SEND_INLINE = (1<<3),
1334 IB_SEND_IP_CSUM = (1<<4),
1336 /* reserve bits 26-31 for low level drivers' internal use */
1337 IB_SEND_RESERVED_START = (1 << 26),
1338 IB_SEND_RESERVED_END = (1 << 31),
1348 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1352 struct ib_send_wr *next;
1355 struct ib_cqe *wr_cqe;
1357 struct ib_sge *sg_list;
1359 enum ib_wr_opcode opcode;
1363 u32 invalidate_rkey;
1368 struct ib_send_wr wr;
1373 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1375 return container_of(wr, struct ib_rdma_wr, wr);
1378 struct ib_atomic_wr {
1379 struct ib_send_wr wr;
1383 u64 compare_add_mask;
1388 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1390 return container_of(wr, struct ib_atomic_wr, wr);
1394 struct ib_send_wr wr;
1401 u16 pkey_index; /* valid for GSI only */
1402 u8 port_num; /* valid for DR SMPs on switch only */
1405 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1407 return container_of(wr, struct ib_ud_wr, wr);
1411 struct ib_send_wr wr;
1417 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1419 return container_of(wr, struct ib_reg_wr, wr);
1422 struct ib_sig_handover_wr {
1423 struct ib_send_wr wr;
1424 struct ib_sig_attrs *sig_attrs;
1425 struct ib_mr *sig_mr;
1427 struct ib_sge *prot;
1430 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1432 return container_of(wr, struct ib_sig_handover_wr, wr);
1436 struct ib_recv_wr *next;
1439 struct ib_cqe *wr_cqe;
1441 struct ib_sge *sg_list;
1445 enum ib_access_flags {
1446 IB_ACCESS_LOCAL_WRITE = 1,
1447 IB_ACCESS_REMOTE_WRITE = (1<<1),
1448 IB_ACCESS_REMOTE_READ = (1<<2),
1449 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1450 IB_ACCESS_MW_BIND = (1<<4),
1451 IB_ZERO_BASED = (1<<5),
1452 IB_ACCESS_ON_DEMAND = (1<<6),
1453 IB_ACCESS_HUGETLB = (1<<7),
1457 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1458 * are hidden here instead of a uapi header!
1460 enum ib_mr_rereg_flags {
1461 IB_MR_REREG_TRANS = 1,
1462 IB_MR_REREG_PD = (1<<1),
1463 IB_MR_REREG_ACCESS = (1<<2),
1464 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1467 struct ib_fmr_attr {
1475 enum rdma_remove_reason {
1476 /* Userspace requested uobject deletion. Call could fail */
1477 RDMA_REMOVE_DESTROY,
1478 /* Context deletion. This call should delete the actual object itself */
1480 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1481 RDMA_REMOVE_DRIVER_REMOVE,
1482 /* Context is being cleaned-up, but commit was just completed */
1483 RDMA_REMOVE_DURING_CLEANUP,
1486 struct ib_rdmacg_object {
1487 #ifdef CONFIG_CGROUP_RDMA
1488 struct rdma_cgroup *cg; /* owner rdma cgroup */
1492 struct ib_ucontext {
1493 struct ib_device *device;
1494 struct ib_uverbs_file *ufile;
1497 /* locking the uobjects_list */
1498 struct mutex uobjects_lock;
1499 struct list_head uobjects;
1500 /* protects cleanup process from other actions */
1501 struct rw_semaphore cleanup_rwsem;
1502 enum rdma_remove_reason cleanup_reason;
1505 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1506 struct rb_root_cached umem_tree;
1508 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1509 * mmu notifiers registration.
1511 struct rw_semaphore umem_rwsem;
1512 void (*invalidate_range)(struct ib_umem *umem,
1513 unsigned long start, unsigned long end);
1515 struct mmu_notifier mn;
1516 atomic_t notifier_count;
1517 /* A list of umems that don't have private mmu notifier counters yet. */
1518 struct list_head no_private_counters;
1522 struct ib_rdmacg_object cg_obj;
1526 u64 user_handle; /* handle given to us by userspace */
1527 struct ib_ucontext *context; /* associated user context */
1528 void *object; /* containing object */
1529 struct list_head list; /* link to context's list */
1530 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1531 int id; /* index into kernel idr */
1533 atomic_t usecnt; /* protects exclusive access */
1534 struct rcu_head rcu; /* kfree_rcu() overhead */
1536 const struct uverbs_obj_type *type;
1539 struct ib_uobject_file {
1540 struct ib_uobject uobj;
1541 /* ufile contains the lock between context release and file close */
1542 struct ib_uverbs_file *ufile;
1546 const void __user *inbuf;
1547 void __user *outbuf;
1555 struct ib_device *device;
1556 struct ib_uobject *uobject;
1557 atomic_t usecnt; /* count all resources */
1559 u32 unsafe_global_rkey;
1562 * Implementation details of the RDMA core, don't use in drivers:
1564 struct ib_mr *__internal_mr;
1565 struct rdma_restrack_entry res;
1569 struct ib_device *device;
1570 atomic_t usecnt; /* count all exposed resources */
1571 struct inode *inode;
1573 struct mutex tgt_qp_mutex;
1574 struct list_head tgt_qp_list;
1578 struct ib_device *device;
1580 struct ib_uobject *uobject;
1581 enum rdma_ah_attr_type type;
1584 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1586 enum ib_poll_context {
1587 IB_POLL_DIRECT, /* caller context, no hw completions */
1588 IB_POLL_SOFTIRQ, /* poll from softirq context */
1589 IB_POLL_WORKQUEUE, /* poll from workqueue */
1593 struct ib_device *device;
1594 struct ib_uobject *uobject;
1595 ib_comp_handler comp_handler;
1596 void (*event_handler)(struct ib_event *, void *);
1599 atomic_t usecnt; /* count number of work queues */
1600 enum ib_poll_context poll_ctx;
1603 struct irq_poll iop;
1604 struct work_struct work;
1607 * Implementation details of the RDMA core, don't use in drivers:
1609 struct rdma_restrack_entry res;
1613 struct ib_device *device;
1615 struct ib_uobject *uobject;
1616 void (*event_handler)(struct ib_event *, void *);
1618 enum ib_srq_type srq_type;
1625 struct ib_xrcd *xrcd;
1632 enum ib_raw_packet_caps {
1633 /* Strip cvlan from incoming packet and report it in the matching work
1634 * completion is supported.
1636 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1637 /* Scatter FCS field of an incoming packet to host memory is supported.
1639 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1640 /* Checksum offloads are supported (for both send and receive). */
1641 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1642 /* When a packet is received for an RQ with no receive WQEs, the
1643 * packet processing is delayed.
1645 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1659 struct ib_device *device;
1660 struct ib_uobject *uobject;
1662 void (*event_handler)(struct ib_event *, void *);
1666 enum ib_wq_state state;
1667 enum ib_wq_type wq_type;
1672 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1673 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1674 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1675 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1678 struct ib_wq_init_attr {
1680 enum ib_wq_type wq_type;
1684 void (*event_handler)(struct ib_event *, void *);
1685 u32 create_flags; /* Use enum ib_wq_flags */
1688 enum ib_wq_attr_mask {
1689 IB_WQ_STATE = 1 << 0,
1690 IB_WQ_CUR_STATE = 1 << 1,
1691 IB_WQ_FLAGS = 1 << 2,
1695 enum ib_wq_state wq_state;
1696 enum ib_wq_state curr_wq_state;
1697 u32 flags; /* Use enum ib_wq_flags */
1698 u32 flags_mask; /* Use enum ib_wq_flags */
1701 struct ib_rwq_ind_table {
1702 struct ib_device *device;
1703 struct ib_uobject *uobject;
1706 u32 log_ind_tbl_size;
1707 struct ib_wq **ind_tbl;
1710 struct ib_rwq_ind_table_init_attr {
1711 u32 log_ind_tbl_size;
1712 /* Each entry is a pointer to Receive Work Queue */
1713 struct ib_wq **ind_tbl;
1716 enum port_pkey_state {
1717 IB_PORT_PKEY_NOT_VALID = 0,
1718 IB_PORT_PKEY_VALID = 1,
1719 IB_PORT_PKEY_LISTED = 2,
1722 struct ib_qp_security;
1724 struct ib_port_pkey {
1725 enum port_pkey_state state;
1728 struct list_head qp_list;
1729 struct list_head to_error_list;
1730 struct ib_qp_security *sec;
1733 struct ib_ports_pkeys {
1734 struct ib_port_pkey main;
1735 struct ib_port_pkey alt;
1738 struct ib_qp_security {
1740 struct ib_device *dev;
1741 /* Hold this mutex when changing port and pkey settings. */
1743 struct ib_ports_pkeys *ports_pkeys;
1744 /* A list of all open shared QP handles. Required to enforce security
1745 * properly for all users of a shared QP.
1747 struct list_head shared_qp_list;
1750 atomic_t error_list_count;
1751 struct completion error_complete;
1752 int error_comps_pending;
1756 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1757 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1760 struct ib_device *device;
1762 struct ib_cq *send_cq;
1763 struct ib_cq *recv_cq;
1766 struct list_head rdma_mrs;
1767 struct list_head sig_mrs;
1769 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1770 struct list_head xrcd_list;
1772 /* count times opened, mcast attaches, flow attaches */
1774 struct list_head open_list;
1775 struct ib_qp *real_qp;
1776 struct ib_uobject *uobject;
1777 void (*event_handler)(struct ib_event *, void *);
1782 enum ib_qp_type qp_type;
1783 struct ib_rwq_ind_table *rwq_ind_tbl;
1784 struct ib_qp_security *qp_sec;
1788 * Implementation details of the RDMA core, don't use in drivers:
1790 struct rdma_restrack_entry res;
1794 struct ib_device *device;
1797 struct ib_uobject *uobject;
1802 struct ib_device *device;
1808 unsigned int page_size;
1811 struct ib_uobject *uobject; /* user */
1812 struct list_head qp_entry; /* FR */
1818 * Implementation details of the RDMA core, don't use in drivers:
1820 struct rdma_restrack_entry res;
1824 struct ib_device *device;
1826 struct ib_uobject *uobject;
1828 enum ib_mw_type type;
1832 struct ib_device *device;
1834 struct list_head list;
1839 /* Supported steering options */
1840 enum ib_flow_attr_type {
1841 /* steering according to rule specifications */
1842 IB_FLOW_ATTR_NORMAL = 0x0,
1843 /* default unicast and multicast rule -
1844 * receive all Eth traffic which isn't steered to any QP
1846 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1847 /* default multicast rule -
1848 * receive all Eth multicast traffic which isn't steered to any QP
1850 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1851 /* sniffer rule - receive all port traffic */
1852 IB_FLOW_ATTR_SNIFFER = 0x3
1855 /* Supported steering header types */
1856 enum ib_flow_spec_type {
1858 IB_FLOW_SPEC_ETH = 0x20,
1859 IB_FLOW_SPEC_IB = 0x22,
1861 IB_FLOW_SPEC_IPV4 = 0x30,
1862 IB_FLOW_SPEC_IPV6 = 0x31,
1863 IB_FLOW_SPEC_ESP = 0x34,
1865 IB_FLOW_SPEC_TCP = 0x40,
1866 IB_FLOW_SPEC_UDP = 0x41,
1867 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1868 IB_FLOW_SPEC_GRE = 0x51,
1869 IB_FLOW_SPEC_MPLS = 0x60,
1870 IB_FLOW_SPEC_INNER = 0x100,
1872 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1873 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1874 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1875 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1877 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1878 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1880 /* Flow steering rule priority is set according to it's domain.
1881 * Lower domain value means higher priority.
1883 enum ib_flow_domain {
1884 IB_FLOW_DOMAIN_USER,
1885 IB_FLOW_DOMAIN_ETHTOOL,
1888 IB_FLOW_DOMAIN_NUM /* Must be last */
1891 enum ib_flow_flags {
1892 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1893 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1894 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1897 struct ib_flow_eth_filter {
1906 struct ib_flow_spec_eth {
1909 struct ib_flow_eth_filter val;
1910 struct ib_flow_eth_filter mask;
1913 struct ib_flow_ib_filter {
1920 struct ib_flow_spec_ib {
1923 struct ib_flow_ib_filter val;
1924 struct ib_flow_ib_filter mask;
1927 /* IPv4 header flags */
1928 enum ib_ipv4_flags {
1929 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1930 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1931 last have this flag set */
1934 struct ib_flow_ipv4_filter {
1945 struct ib_flow_spec_ipv4 {
1948 struct ib_flow_ipv4_filter val;
1949 struct ib_flow_ipv4_filter mask;
1952 struct ib_flow_ipv6_filter {
1963 struct ib_flow_spec_ipv6 {
1966 struct ib_flow_ipv6_filter val;
1967 struct ib_flow_ipv6_filter mask;
1970 struct ib_flow_tcp_udp_filter {
1977 struct ib_flow_spec_tcp_udp {
1980 struct ib_flow_tcp_udp_filter val;
1981 struct ib_flow_tcp_udp_filter mask;
1984 struct ib_flow_tunnel_filter {
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990 * the tunnel_id from val has the vni value
1992 struct ib_flow_spec_tunnel {
1995 struct ib_flow_tunnel_filter val;
1996 struct ib_flow_tunnel_filter mask;
1999 struct ib_flow_esp_filter {
2006 struct ib_flow_spec_esp {
2009 struct ib_flow_esp_filter val;
2010 struct ib_flow_esp_filter mask;
2013 struct ib_flow_gre_filter {
2014 __be16 c_ks_res0_ver;
2021 struct ib_flow_spec_gre {
2024 struct ib_flow_gre_filter val;
2025 struct ib_flow_gre_filter mask;
2028 struct ib_flow_mpls_filter {
2034 struct ib_flow_spec_mpls {
2037 struct ib_flow_mpls_filter val;
2038 struct ib_flow_mpls_filter mask;
2041 struct ib_flow_spec_action_tag {
2042 enum ib_flow_spec_type type;
2047 struct ib_flow_spec_action_drop {
2048 enum ib_flow_spec_type type;
2052 struct ib_flow_spec_action_handle {
2053 enum ib_flow_spec_type type;
2055 struct ib_flow_action *act;
2058 enum ib_counters_description {
2063 struct ib_flow_spec_action_count {
2064 enum ib_flow_spec_type type;
2066 struct ib_counters *counters;
2069 union ib_flow_spec {
2074 struct ib_flow_spec_eth eth;
2075 struct ib_flow_spec_ib ib;
2076 struct ib_flow_spec_ipv4 ipv4;
2077 struct ib_flow_spec_tcp_udp tcp_udp;
2078 struct ib_flow_spec_ipv6 ipv6;
2079 struct ib_flow_spec_tunnel tunnel;
2080 struct ib_flow_spec_esp esp;
2081 struct ib_flow_spec_gre gre;
2082 struct ib_flow_spec_mpls mpls;
2083 struct ib_flow_spec_action_tag flow_tag;
2084 struct ib_flow_spec_action_drop drop;
2085 struct ib_flow_spec_action_handle action;
2086 struct ib_flow_spec_action_count flow_count;
2089 struct ib_flow_attr {
2090 enum ib_flow_attr_type type;
2096 /* Following are the optional layers according to user request
2097 * struct ib_flow_spec_xxx
2098 * struct ib_flow_spec_yyy
2104 struct ib_uobject *uobject;
2107 enum ib_flow_action_type {
2108 IB_FLOW_ACTION_UNSPECIFIED,
2109 IB_FLOW_ACTION_ESP = 1,
2112 struct ib_flow_action_attrs_esp_keymats {
2113 enum ib_uverbs_flow_action_esp_keymat protocol;
2115 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2119 struct ib_flow_action_attrs_esp_replays {
2120 enum ib_uverbs_flow_action_esp_replay protocol;
2122 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2126 enum ib_flow_action_attrs_esp_flags {
2127 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2128 * This is done in order to share the same flags between user-space and
2129 * kernel and spare an unnecessary translation.
2133 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2134 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2137 struct ib_flow_spec_list {
2138 struct ib_flow_spec_list *next;
2139 union ib_flow_spec spec;
2142 struct ib_flow_action_attrs_esp {
2143 struct ib_flow_action_attrs_esp_keymats *keymat;
2144 struct ib_flow_action_attrs_esp_replays *replay;
2145 struct ib_flow_spec_list *encap;
2146 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2147 * Value of 0 is a valid value.
2153 /* Use enum ib_flow_action_attrs_esp_flags */
2155 u64 hard_limit_pkts;
2158 struct ib_flow_action {
2159 struct ib_device *device;
2160 struct ib_uobject *uobject;
2161 enum ib_flow_action_type type;
2168 enum ib_process_mad_flags {
2169 IB_MAD_IGNORE_MKEY = 1,
2170 IB_MAD_IGNORE_BKEY = 2,
2171 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2174 enum ib_mad_result {
2175 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2176 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2177 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2178 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2181 struct ib_port_cache {
2183 struct ib_pkey_cache *pkey;
2184 struct ib_gid_table *gid;
2186 enum ib_port_state port_state;
2191 struct ib_event_handler event_handler;
2192 struct ib_port_cache *ports;
2197 struct ib_port_immutable {
2204 /* rdma netdev type - specifies protocol type */
2205 enum rdma_netdev_t {
2206 RDMA_NETDEV_OPA_VNIC,
2211 * struct rdma_netdev - rdma netdev
2212 * For cases where netstack interfacing is required.
2214 struct rdma_netdev {
2216 struct ib_device *hca;
2219 /* cleanup function must be specified */
2220 void (*free_rdma_netdev)(struct net_device *netdev);
2222 /* control functions */
2223 void (*set_id)(struct net_device *netdev, int id);
2225 int (*send)(struct net_device *dev, struct sk_buff *skb,
2226 struct ib_ah *address, u32 dqpn);
2228 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2229 union ib_gid *gid, u16 mlid,
2230 int set_qkey, u32 qkey);
2231 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2232 union ib_gid *gid, u16 mlid);
2235 struct ib_port_pkey_list {
2236 /* Lock to hold while modifying the list. */
2237 spinlock_t list_lock;
2238 struct list_head pkey_list;
2241 struct ib_counters {
2242 struct ib_device *device;
2243 struct ib_uobject *uobject;
2244 /* num of objects attached */
2248 enum ib_read_counters_flags {
2249 /* prefer read values from driver cache */
2250 IB_READ_COUNTERS_ATTR_PREFER_CACHED = 1 << 0,
2253 struct ib_counters_read_attr {
2256 u32 flags; /* use enum ib_read_counters_flags */
2259 struct uverbs_attr_bundle;
2262 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2263 struct device *dma_device;
2265 char name[IB_DEVICE_NAME_MAX];
2267 struct list_head event_handler_list;
2268 spinlock_t event_handler_lock;
2270 spinlock_t client_data_lock;
2271 struct list_head core_list;
2272 /* Access to the client_data_list is protected by the client_data_lock
2273 * spinlock and the lists_rwsem read-write semaphore */
2274 struct list_head client_data_list;
2276 struct ib_cache cache;
2278 * port_immutable is indexed by port number
2280 struct ib_port_immutable *port_immutable;
2282 int num_comp_vectors;
2284 struct ib_port_pkey_list *port_pkey_list;
2286 struct iw_cm_verbs *iwcm;
2289 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2290 * driver initialized data. The struct is kfree()'ed by the sysfs
2291 * core when the device is removed. A lifespan of -1 in the return
2292 * struct tells the core to set a default lifespan.
2294 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2297 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2298 * @index - The index in the value array we wish to have updated, or
2299 * num_counters if we want all stats updated
2301 * < 0 - Error, no counters updated
2302 * index - Updated the single counter pointed to by index
2303 * num_counters - Updated all counters (will reset the timestamp
2304 * and prevent further calls for lifespan milliseconds)
2305 * Drivers are allowed to update all counters in leiu of just the
2306 * one given in index at their option
2308 int (*get_hw_stats)(struct ib_device *device,
2309 struct rdma_hw_stats *stats,
2310 u8 port, int index);
2311 int (*query_device)(struct ib_device *device,
2312 struct ib_device_attr *device_attr,
2313 struct ib_udata *udata);
2314 int (*query_port)(struct ib_device *device,
2316 struct ib_port_attr *port_attr);
2317 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2319 /* When calling get_netdev, the HW vendor's driver should return the
2320 * net device of device @device at port @port_num or NULL if such
2321 * a net device doesn't exist. The vendor driver should call dev_hold
2322 * on this net device. The HW vendor's device driver must guarantee
2323 * that this function returns NULL before the net device has finished
2324 * NETDEV_UNREGISTER state.
2326 struct net_device *(*get_netdev)(struct ib_device *device,
2328 /* query_gid should be return GID value for @device, when @port_num
2329 * link layer is either IB or iWarp. It is no-op if @port_num port
2330 * is RoCE link layer.
2332 int (*query_gid)(struct ib_device *device,
2333 u8 port_num, int index,
2335 /* When calling add_gid, the HW vendor's driver should add the gid
2336 * of device of port at gid index available at @attr. Meta-info of
2337 * that gid (for example, the network device related to this gid) is
2338 * available at @attr. @context allows the HW vendor driver to store
2339 * extra information together with a GID entry. The HW vendor driver may
2340 * allocate memory to contain this information and store it in @context
2341 * when a new GID entry is written to. Params are consistent until the
2342 * next call of add_gid or delete_gid. The function should return 0 on
2343 * success or error otherwise. The function could be called
2344 * concurrently for different ports. This function is only called when
2345 * roce_gid_table is used.
2347 int (*add_gid)(const union ib_gid *gid,
2348 const struct ib_gid_attr *attr,
2350 /* When calling del_gid, the HW vendor's driver should delete the
2351 * gid of device @device at gid index gid_index of port port_num
2352 * available in @attr.
2353 * Upon the deletion of a GID entry, the HW vendor must free any
2354 * allocated memory. The caller will clear @context afterwards.
2355 * This function is only called when roce_gid_table is used.
2357 int (*del_gid)(const struct ib_gid_attr *attr,
2359 int (*query_pkey)(struct ib_device *device,
2360 u8 port_num, u16 index, u16 *pkey);
2361 int (*modify_device)(struct ib_device *device,
2362 int device_modify_mask,
2363 struct ib_device_modify *device_modify);
2364 int (*modify_port)(struct ib_device *device,
2365 u8 port_num, int port_modify_mask,
2366 struct ib_port_modify *port_modify);
2367 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2368 struct ib_udata *udata);
2369 int (*dealloc_ucontext)(struct ib_ucontext *context);
2370 int (*mmap)(struct ib_ucontext *context,
2371 struct vm_area_struct *vma);
2372 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2373 struct ib_ucontext *context,
2374 struct ib_udata *udata);
2375 int (*dealloc_pd)(struct ib_pd *pd);
2376 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2377 struct rdma_ah_attr *ah_attr,
2378 struct ib_udata *udata);
2379 int (*modify_ah)(struct ib_ah *ah,
2380 struct rdma_ah_attr *ah_attr);
2381 int (*query_ah)(struct ib_ah *ah,
2382 struct rdma_ah_attr *ah_attr);
2383 int (*destroy_ah)(struct ib_ah *ah);
2384 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2385 struct ib_srq_init_attr *srq_init_attr,
2386 struct ib_udata *udata);
2387 int (*modify_srq)(struct ib_srq *srq,
2388 struct ib_srq_attr *srq_attr,
2389 enum ib_srq_attr_mask srq_attr_mask,
2390 struct ib_udata *udata);
2391 int (*query_srq)(struct ib_srq *srq,
2392 struct ib_srq_attr *srq_attr);
2393 int (*destroy_srq)(struct ib_srq *srq);
2394 int (*post_srq_recv)(struct ib_srq *srq,
2395 struct ib_recv_wr *recv_wr,
2396 struct ib_recv_wr **bad_recv_wr);
2397 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2398 struct ib_qp_init_attr *qp_init_attr,
2399 struct ib_udata *udata);
2400 int (*modify_qp)(struct ib_qp *qp,
2401 struct ib_qp_attr *qp_attr,
2403 struct ib_udata *udata);
2404 int (*query_qp)(struct ib_qp *qp,
2405 struct ib_qp_attr *qp_attr,
2407 struct ib_qp_init_attr *qp_init_attr);
2408 int (*destroy_qp)(struct ib_qp *qp);
2409 int (*post_send)(struct ib_qp *qp,
2410 struct ib_send_wr *send_wr,
2411 struct ib_send_wr **bad_send_wr);
2412 int (*post_recv)(struct ib_qp *qp,
2413 struct ib_recv_wr *recv_wr,
2414 struct ib_recv_wr **bad_recv_wr);
2415 struct ib_cq * (*create_cq)(struct ib_device *device,
2416 const struct ib_cq_init_attr *attr,
2417 struct ib_ucontext *context,
2418 struct ib_udata *udata);
2419 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2421 int (*destroy_cq)(struct ib_cq *cq);
2422 int (*resize_cq)(struct ib_cq *cq, int cqe,
2423 struct ib_udata *udata);
2424 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2426 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2427 int (*req_notify_cq)(struct ib_cq *cq,
2428 enum ib_cq_notify_flags flags);
2429 int (*req_ncomp_notif)(struct ib_cq *cq,
2431 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2432 int mr_access_flags);
2433 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2434 u64 start, u64 length,
2436 int mr_access_flags,
2437 struct ib_udata *udata);
2438 int (*rereg_user_mr)(struct ib_mr *mr,
2440 u64 start, u64 length,
2442 int mr_access_flags,
2444 struct ib_udata *udata);
2445 int (*dereg_mr)(struct ib_mr *mr);
2446 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2447 enum ib_mr_type mr_type,
2449 int (*map_mr_sg)(struct ib_mr *mr,
2450 struct scatterlist *sg,
2452 unsigned int *sg_offset);
2453 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2454 enum ib_mw_type type,
2455 struct ib_udata *udata);
2456 int (*dealloc_mw)(struct ib_mw *mw);
2457 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2458 int mr_access_flags,
2459 struct ib_fmr_attr *fmr_attr);
2460 int (*map_phys_fmr)(struct ib_fmr *fmr,
2461 u64 *page_list, int list_len,
2463 int (*unmap_fmr)(struct list_head *fmr_list);
2464 int (*dealloc_fmr)(struct ib_fmr *fmr);
2465 int (*attach_mcast)(struct ib_qp *qp,
2468 int (*detach_mcast)(struct ib_qp *qp,
2471 int (*process_mad)(struct ib_device *device,
2472 int process_mad_flags,
2474 const struct ib_wc *in_wc,
2475 const struct ib_grh *in_grh,
2476 const struct ib_mad_hdr *in_mad,
2478 struct ib_mad_hdr *out_mad,
2479 size_t *out_mad_size,
2480 u16 *out_mad_pkey_index);
2481 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2482 struct ib_ucontext *ucontext,
2483 struct ib_udata *udata);
2484 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2485 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2489 struct ib_udata *udata);
2490 int (*destroy_flow)(struct ib_flow *flow_id);
2491 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2492 struct ib_mr_status *mr_status);
2493 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2494 void (*drain_rq)(struct ib_qp *qp);
2495 void (*drain_sq)(struct ib_qp *qp);
2496 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2498 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2499 struct ifla_vf_info *ivf);
2500 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2501 struct ifla_vf_stats *stats);
2502 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2504 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2505 struct ib_wq_init_attr *init_attr,
2506 struct ib_udata *udata);
2507 int (*destroy_wq)(struct ib_wq *wq);
2508 int (*modify_wq)(struct ib_wq *wq,
2509 struct ib_wq_attr *attr,
2511 struct ib_udata *udata);
2512 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2513 struct ib_rwq_ind_table_init_attr *init_attr,
2514 struct ib_udata *udata);
2515 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2516 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2517 const struct ib_flow_action_attrs_esp *attr,
2518 struct uverbs_attr_bundle *attrs);
2519 int (*destroy_flow_action)(struct ib_flow_action *action);
2520 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2521 const struct ib_flow_action_attrs_esp *attr,
2522 struct uverbs_attr_bundle *attrs);
2523 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2524 struct ib_ucontext *context,
2525 struct ib_dm_alloc_attr *attr,
2526 struct uverbs_attr_bundle *attrs);
2527 int (*dealloc_dm)(struct ib_dm *dm);
2528 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2529 struct ib_dm_mr_attr *attr,
2530 struct uverbs_attr_bundle *attrs);
2531 struct ib_counters * (*create_counters)(struct ib_device *device,
2532 struct uverbs_attr_bundle *attrs);
2533 int (*destroy_counters)(struct ib_counters *counters);
2534 int (*read_counters)(struct ib_counters *counters,
2535 struct ib_counters_read_attr *counters_read_attr,
2536 struct uverbs_attr_bundle *attrs);
2539 * rdma netdev operation
2541 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2542 * doesn't support the specified rdma netdev type.
2544 struct net_device *(*alloc_rdma_netdev)(
2545 struct ib_device *device,
2547 enum rdma_netdev_t type,
2549 unsigned char name_assign_type,
2550 void (*setup)(struct net_device *));
2552 struct module *owner;
2554 struct kobject *ports_parent;
2555 struct list_head port_list;
2558 IB_DEV_UNINITIALIZED,
2564 u64 uverbs_cmd_mask;
2565 u64 uverbs_ex_cmd_mask;
2567 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2573 struct ib_device_attr attrs;
2574 struct attribute_group *hw_stats_ag;
2575 struct rdma_hw_stats *hw_stats;
2577 #ifdef CONFIG_CGROUP_RDMA
2578 struct rdmacg_device cg_device;
2583 * Implementation details of the RDMA core, don't use in drivers
2585 struct rdma_restrack_root res;
2588 * The following mandatory functions are used only at device
2589 * registration. Keep functions such as these at the end of this
2590 * structure to avoid cache line misses when accessing struct ib_device
2593 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2594 void (*get_dev_fw_str)(struct ib_device *, char *str);
2595 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2598 struct uverbs_root_spec *specs_root;
2599 enum rdma_driver_id driver_id;
2604 void (*add) (struct ib_device *);
2605 void (*remove)(struct ib_device *, void *client_data);
2607 /* Returns the net_dev belonging to this ib_client and matching the
2609 * @dev: An RDMA device that the net_dev use for communication.
2610 * @port: A physical port number on the RDMA device.
2611 * @pkey: P_Key that the net_dev uses if applicable.
2612 * @gid: A GID that the net_dev uses to communicate.
2613 * @addr: An IP address the net_dev is configured with.
2614 * @client_data: The device's client data set by ib_set_client_data().
2616 * An ib_client that implements a net_dev on top of RDMA devices
2617 * (such as IP over IB) should implement this callback, allowing the
2618 * rdma_cm module to find the right net_dev for a given request.
2620 * The caller is responsible for calling dev_put on the returned
2622 struct net_device *(*get_net_dev_by_params)(
2623 struct ib_device *dev,
2626 const union ib_gid *gid,
2627 const struct sockaddr *addr,
2629 struct list_head list;
2632 struct ib_device *ib_alloc_device(size_t size);
2633 void ib_dealloc_device(struct ib_device *device);
2635 void ib_get_device_fw_str(struct ib_device *device, char *str);
2637 int ib_register_device(struct ib_device *device,
2638 int (*port_callback)(struct ib_device *,
2639 u8, struct kobject *));
2640 void ib_unregister_device(struct ib_device *device);
2642 int ib_register_client (struct ib_client *client);
2643 void ib_unregister_client(struct ib_client *client);
2645 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2646 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2649 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2651 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2654 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2656 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2659 static inline bool ib_is_buffer_cleared(const void __user *p,
2665 if (len > USHRT_MAX)
2668 buf = memdup_user(p, len);
2672 ret = !memchr_inv(buf, 0, len);
2677 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2681 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2685 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2686 * contains all required attributes and no attributes not allowed for
2687 * the given QP state transition.
2688 * @cur_state: Current QP state
2689 * @next_state: Next QP state
2691 * @mask: Mask of supplied QP attributes
2692 * @ll : link layer of port
2694 * This function is a helper function that a low-level driver's
2695 * modify_qp method can use to validate the consumer's input. It
2696 * checks that cur_state and next_state are valid QP states, that a
2697 * transition from cur_state to next_state is allowed by the IB spec,
2698 * and that the attribute mask supplied is allowed for the transition.
2700 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2701 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2702 enum rdma_link_layer ll);
2704 void ib_register_event_handler(struct ib_event_handler *event_handler);
2705 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2706 void ib_dispatch_event(struct ib_event *event);
2708 int ib_query_port(struct ib_device *device,
2709 u8 port_num, struct ib_port_attr *port_attr);
2711 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2715 * rdma_cap_ib_switch - Check if the device is IB switch
2716 * @device: Device to check
2718 * Device driver is responsible for setting is_switch bit on
2719 * in ib_device structure at init time.
2721 * Return: true if the device is IB switch.
2723 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2725 return device->is_switch;
2729 * rdma_start_port - Return the first valid port number for the device
2732 * @device: Device to be checked
2734 * Return start port number
2736 static inline u8 rdma_start_port(const struct ib_device *device)
2738 return rdma_cap_ib_switch(device) ? 0 : 1;
2742 * rdma_end_port - Return the last valid port number for the device
2745 * @device: Device to be checked
2747 * Return last port number
2749 static inline u8 rdma_end_port(const struct ib_device *device)
2751 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2754 static inline int rdma_is_port_valid(const struct ib_device *device,
2757 return (port >= rdma_start_port(device) &&
2758 port <= rdma_end_port(device));
2761 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2763 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2766 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2768 return device->port_immutable[port_num].core_cap_flags &
2769 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2772 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2774 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2777 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2779 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2782 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2784 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2787 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2789 return rdma_protocol_ib(device, port_num) ||
2790 rdma_protocol_roce(device, port_num);
2793 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2795 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2798 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2800 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2804 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2805 * Management Datagrams.
2806 * @device: Device to check
2807 * @port_num: Port number to check
2809 * Management Datagrams (MAD) are a required part of the InfiniBand
2810 * specification and are supported on all InfiniBand devices. A slightly
2811 * extended version are also supported on OPA interfaces.
2813 * Return: true if the port supports sending/receiving of MAD packets.
2815 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2817 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2821 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2822 * Management Datagrams.
2823 * @device: Device to check
2824 * @port_num: Port number to check
2826 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2827 * datagrams with their own versions. These OPA MADs share many but not all of
2828 * the characteristics of InfiniBand MADs.
2830 * OPA MADs differ in the following ways:
2832 * 1) MADs are variable size up to 2K
2833 * IBTA defined MADs remain fixed at 256 bytes
2834 * 2) OPA SMPs must carry valid PKeys
2835 * 3) OPA SMP packets are a different format
2837 * Return: true if the port supports OPA MAD packet formats.
2839 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2841 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2842 == RDMA_CORE_CAP_OPA_MAD;
2846 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2847 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2848 * @device: Device to check
2849 * @port_num: Port number to check
2851 * Each InfiniBand node is required to provide a Subnet Management Agent
2852 * that the subnet manager can access. Prior to the fabric being fully
2853 * configured by the subnet manager, the SMA is accessed via a well known
2854 * interface called the Subnet Management Interface (SMI). This interface
2855 * uses directed route packets to communicate with the SM to get around the
2856 * chicken and egg problem of the SM needing to know what's on the fabric
2857 * in order to configure the fabric, and needing to configure the fabric in
2858 * order to send packets to the devices on the fabric. These directed
2859 * route packets do not need the fabric fully configured in order to reach
2860 * their destination. The SMI is the only method allowed to send
2861 * directed route packets on an InfiniBand fabric.
2863 * Return: true if the port provides an SMI.
2865 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2867 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2871 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2872 * Communication Manager.
2873 * @device: Device to check
2874 * @port_num: Port number to check
2876 * The InfiniBand Communication Manager is one of many pre-defined General
2877 * Service Agents (GSA) that are accessed via the General Service
2878 * Interface (GSI). It's role is to facilitate establishment of connections
2879 * between nodes as well as other management related tasks for established
2882 * Return: true if the port supports an IB CM (this does not guarantee that
2883 * a CM is actually running however).
2885 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2887 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2891 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2892 * Communication Manager.
2893 * @device: Device to check
2894 * @port_num: Port number to check
2896 * Similar to above, but specific to iWARP connections which have a different
2897 * managment protocol than InfiniBand.
2899 * Return: true if the port supports an iWARP CM (this does not guarantee that
2900 * a CM is actually running however).
2902 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2904 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2908 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2909 * Subnet Administration.
2910 * @device: Device to check
2911 * @port_num: Port number to check
2913 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2914 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2915 * fabrics, devices should resolve routes to other hosts by contacting the
2916 * SA to query the proper route.
2918 * Return: true if the port should act as a client to the fabric Subnet
2919 * Administration interface. This does not imply that the SA service is
2922 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2924 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2928 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2930 * @device: Device to check
2931 * @port_num: Port number to check
2933 * InfiniBand multicast registration is more complex than normal IPv4 or
2934 * IPv6 multicast registration. Each Host Channel Adapter must register
2935 * with the Subnet Manager when it wishes to join a multicast group. It
2936 * should do so only once regardless of how many queue pairs it subscribes
2937 * to this group. And it should leave the group only after all queue pairs
2938 * attached to the group have been detached.
2940 * Return: true if the port must undertake the additional adminstrative
2941 * overhead of registering/unregistering with the SM and tracking of the
2942 * total number of queue pairs attached to the multicast group.
2944 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2946 return rdma_cap_ib_sa(device, port_num);
2950 * rdma_cap_af_ib - Check if the port of device has the capability
2951 * Native Infiniband Address.
2952 * @device: Device to check
2953 * @port_num: Port number to check
2955 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2956 * GID. RoCE uses a different mechanism, but still generates a GID via
2957 * a prescribed mechanism and port specific data.
2959 * Return: true if the port uses a GID address to identify devices on the
2962 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2964 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2968 * rdma_cap_eth_ah - Check if the port of device has the capability
2969 * Ethernet Address Handle.
2970 * @device: Device to check
2971 * @port_num: Port number to check
2973 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2974 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2975 * port. Normally, packet headers are generated by the sending host
2976 * adapter, but when sending connectionless datagrams, we must manually
2977 * inject the proper headers for the fabric we are communicating over.
2979 * Return: true if we are running as a RoCE port and must force the
2980 * addition of a Global Route Header built from our Ethernet Address
2981 * Handle into our header list for connectionless packets.
2983 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2985 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2989 * rdma_cap_opa_ah - Check if the port of device supports
2990 * OPA Address handles
2991 * @device: Device to check
2992 * @port_num: Port number to check
2994 * Return: true if we are running on an OPA device which supports
2995 * the extended OPA addressing.
2997 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2999 return (device->port_immutable[port_num].core_cap_flags &
3000 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3004 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3007 * @port_num: Port number
3009 * This MAD size includes the MAD headers and MAD payload. No other headers
3012 * Return the max MAD size required by the Port. Will return 0 if the port
3013 * does not support MADs
3015 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3017 return device->port_immutable[port_num].max_mad_size;
3021 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3022 * @device: Device to check
3023 * @port_num: Port number to check
3025 * RoCE GID table mechanism manages the various GIDs for a device.
3027 * NOTE: if allocating the port's GID table has failed, this call will still
3028 * return true, but any RoCE GID table API will fail.
3030 * Return: true if the port uses RoCE GID table mechanism in order to manage
3033 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3036 return rdma_protocol_roce(device, port_num) &&
3037 device->add_gid && device->del_gid;
3041 * Check if the device supports READ W/ INVALIDATE.
3043 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3046 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3047 * has support for it yet.
3049 return rdma_protocol_iwarp(dev, port_num);
3052 int ib_query_gid(struct ib_device *device,
3053 u8 port_num, int index, union ib_gid *gid,
3054 struct ib_gid_attr *attr);
3056 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3058 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3059 struct ifla_vf_info *info);
3060 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3061 struct ifla_vf_stats *stats);
3062 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3065 int ib_query_pkey(struct ib_device *device,
3066 u8 port_num, u16 index, u16 *pkey);
3068 int ib_modify_device(struct ib_device *device,
3069 int device_modify_mask,
3070 struct ib_device_modify *device_modify);
3072 int ib_modify_port(struct ib_device *device,
3073 u8 port_num, int port_modify_mask,
3074 struct ib_port_modify *port_modify);
3076 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3077 u8 *port_num, u16 *index);
3079 int ib_find_pkey(struct ib_device *device,
3080 u8 port_num, u16 pkey, u16 *index);
3084 * Create a memory registration for all memory in the system and place
3085 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3086 * ULPs to avoid the overhead of dynamic MRs.
3088 * This flag is generally considered unsafe and must only be used in
3089 * extremly trusted environments. Every use of it will log a warning
3090 * in the kernel log.
3092 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3095 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3096 const char *caller);
3097 #define ib_alloc_pd(device, flags) \
3098 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3099 void ib_dealloc_pd(struct ib_pd *pd);
3102 * rdma_create_ah - Creates an address handle for the given address vector.
3103 * @pd: The protection domain associated with the address handle.
3104 * @ah_attr: The attributes of the address vector.
3106 * The address handle is used to reference a local or global destination
3107 * in all UD QP post sends.
3109 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3112 * rdma_create_user_ah - Creates an address handle for the given address vector.
3113 * It resolves destination mac address for ah attribute of RoCE type.
3114 * @pd: The protection domain associated with the address handle.
3115 * @ah_attr: The attributes of the address vector.
3116 * @udata: pointer to user's input output buffer information need by
3119 * It returns 0 on success and returns appropriate error code on error.
3120 * The address handle is used to reference a local or global destination
3121 * in all UD QP post sends.
3123 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3124 struct rdma_ah_attr *ah_attr,
3125 struct ib_udata *udata);
3127 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3129 * @hdr: the L3 header to parse
3130 * @net_type: type of header to parse
3131 * @sgid: place to store source gid
3132 * @dgid: place to store destination gid
3134 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3135 enum rdma_network_type net_type,
3136 union ib_gid *sgid, union ib_gid *dgid);
3139 * ib_get_rdma_header_version - Get the header version
3140 * @hdr: the L3 header to parse
3142 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3145 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3147 * @device: Device on which the received message arrived.
3148 * @port_num: Port on which the received message arrived.
3149 * @wc: Work completion associated with the received message.
3150 * @grh: References the received global route header. This parameter is
3151 * ignored unless the work completion indicates that the GRH is valid.
3152 * @ah_attr: Returned attributes that can be used when creating an address
3153 * handle for replying to the message.
3155 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3156 const struct ib_wc *wc, const struct ib_grh *grh,
3157 struct rdma_ah_attr *ah_attr);
3160 * ib_create_ah_from_wc - Creates an address handle associated with the
3161 * sender of the specified work completion.
3162 * @pd: The protection domain associated with the address handle.
3163 * @wc: Work completion information associated with a received message.
3164 * @grh: References the received global route header. This parameter is
3165 * ignored unless the work completion indicates that the GRH is valid.
3166 * @port_num: The outbound port number to associate with the address.
3168 * The address handle is used to reference a local or global destination
3169 * in all UD QP post sends.
3171 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3172 const struct ib_grh *grh, u8 port_num);
3175 * rdma_modify_ah - Modifies the address vector associated with an address
3177 * @ah: The address handle to modify.
3178 * @ah_attr: The new address vector attributes to associate with the
3181 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3184 * rdma_query_ah - Queries the address vector associated with an address
3186 * @ah: The address handle to query.
3187 * @ah_attr: The address vector attributes associated with the address
3190 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3193 * rdma_destroy_ah - Destroys an address handle.
3194 * @ah: The address handle to destroy.
3196 int rdma_destroy_ah(struct ib_ah *ah);
3199 * ib_create_srq - Creates a SRQ associated with the specified protection
3201 * @pd: The protection domain associated with the SRQ.
3202 * @srq_init_attr: A list of initial attributes required to create the
3203 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3204 * the actual capabilities of the created SRQ.
3206 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3207 * requested size of the SRQ, and set to the actual values allocated
3208 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3209 * will always be at least as large as the requested values.
3211 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3212 struct ib_srq_init_attr *srq_init_attr);
3215 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3216 * @srq: The SRQ to modify.
3217 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3218 * the current values of selected SRQ attributes are returned.
3219 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3220 * are being modified.
3222 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3223 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3224 * the number of receives queued drops below the limit.
3226 int ib_modify_srq(struct ib_srq *srq,
3227 struct ib_srq_attr *srq_attr,
3228 enum ib_srq_attr_mask srq_attr_mask);
3231 * ib_query_srq - Returns the attribute list and current values for the
3233 * @srq: The SRQ to query.
3234 * @srq_attr: The attributes of the specified SRQ.
3236 int ib_query_srq(struct ib_srq *srq,
3237 struct ib_srq_attr *srq_attr);
3240 * ib_destroy_srq - Destroys the specified SRQ.
3241 * @srq: The SRQ to destroy.
3243 int ib_destroy_srq(struct ib_srq *srq);
3246 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3247 * @srq: The SRQ to post the work request on.
3248 * @recv_wr: A list of work requests to post on the receive queue.
3249 * @bad_recv_wr: On an immediate failure, this parameter will reference
3250 * the work request that failed to be posted on the QP.
3252 static inline int ib_post_srq_recv(struct ib_srq *srq,
3253 struct ib_recv_wr *recv_wr,
3254 struct ib_recv_wr **bad_recv_wr)
3256 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3260 * ib_create_qp - Creates a QP associated with the specified protection
3262 * @pd: The protection domain associated with the QP.
3263 * @qp_init_attr: A list of initial attributes required to create the
3264 * QP. If QP creation succeeds, then the attributes are updated to
3265 * the actual capabilities of the created QP.
3267 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3268 struct ib_qp_init_attr *qp_init_attr);
3271 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3272 * @qp: The QP to modify.
3273 * @attr: On input, specifies the QP attributes to modify. On output,
3274 * the current values of selected QP attributes are returned.
3275 * @attr_mask: A bit-mask used to specify which attributes of the QP
3276 * are being modified.
3277 * @udata: pointer to user's input output buffer information
3278 * are being modified.
3279 * It returns 0 on success and returns appropriate error code on error.
3281 int ib_modify_qp_with_udata(struct ib_qp *qp,
3282 struct ib_qp_attr *attr,
3284 struct ib_udata *udata);
3287 * ib_modify_qp - Modifies the attributes for the specified QP and then
3288 * transitions the QP to the given state.
3289 * @qp: The QP to modify.
3290 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3291 * the current values of selected QP attributes are returned.
3292 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3293 * are being modified.
3295 int ib_modify_qp(struct ib_qp *qp,
3296 struct ib_qp_attr *qp_attr,
3300 * ib_query_qp - Returns the attribute list and current values for the
3302 * @qp: The QP to query.
3303 * @qp_attr: The attributes of the specified QP.
3304 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3305 * @qp_init_attr: Additional attributes of the selected QP.
3307 * The qp_attr_mask may be used to limit the query to gathering only the
3308 * selected attributes.
3310 int ib_query_qp(struct ib_qp *qp,
3311 struct ib_qp_attr *qp_attr,
3313 struct ib_qp_init_attr *qp_init_attr);
3316 * ib_destroy_qp - Destroys the specified QP.
3317 * @qp: The QP to destroy.
3319 int ib_destroy_qp(struct ib_qp *qp);
3322 * ib_open_qp - Obtain a reference to an existing sharable QP.
3323 * @xrcd - XRC domain
3324 * @qp_open_attr: Attributes identifying the QP to open.
3326 * Returns a reference to a sharable QP.
3328 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3329 struct ib_qp_open_attr *qp_open_attr);
3332 * ib_close_qp - Release an external reference to a QP.
3333 * @qp: The QP handle to release
3335 * The opened QP handle is released by the caller. The underlying
3336 * shared QP is not destroyed until all internal references are released.
3338 int ib_close_qp(struct ib_qp *qp);
3341 * ib_post_send - Posts a list of work requests to the send queue of
3343 * @qp: The QP to post the work request on.
3344 * @send_wr: A list of work requests to post on the send queue.
3345 * @bad_send_wr: On an immediate failure, this parameter will reference
3346 * the work request that failed to be posted on the QP.
3348 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3349 * error is returned, the QP state shall not be affected,
3350 * ib_post_send() will return an immediate error after queueing any
3351 * earlier work requests in the list.
3353 static inline int ib_post_send(struct ib_qp *qp,
3354 struct ib_send_wr *send_wr,
3355 struct ib_send_wr **bad_send_wr)
3357 return qp->device->post_send(qp, send_wr, bad_send_wr);
3361 * ib_post_recv - Posts a list of work requests to the receive queue of
3363 * @qp: The QP to post the work request on.
3364 * @recv_wr: A list of work requests to post on the receive queue.
3365 * @bad_recv_wr: On an immediate failure, this parameter will reference
3366 * the work request that failed to be posted on the QP.
3368 static inline int ib_post_recv(struct ib_qp *qp,
3369 struct ib_recv_wr *recv_wr,
3370 struct ib_recv_wr **bad_recv_wr)
3372 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3375 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3376 int nr_cqe, int comp_vector,
3377 enum ib_poll_context poll_ctx, const char *caller);
3378 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3379 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3381 void ib_free_cq(struct ib_cq *cq);
3382 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3385 * ib_create_cq - Creates a CQ on the specified device.
3386 * @device: The device on which to create the CQ.
3387 * @comp_handler: A user-specified callback that is invoked when a
3388 * completion event occurs on the CQ.
3389 * @event_handler: A user-specified callback that is invoked when an
3390 * asynchronous event not associated with a completion occurs on the CQ.
3391 * @cq_context: Context associated with the CQ returned to the user via
3392 * the associated completion and event handlers.
3393 * @cq_attr: The attributes the CQ should be created upon.
3395 * Users can examine the cq structure to determine the actual CQ size.
3397 struct ib_cq *ib_create_cq(struct ib_device *device,
3398 ib_comp_handler comp_handler,
3399 void (*event_handler)(struct ib_event *, void *),
3401 const struct ib_cq_init_attr *cq_attr);
3404 * ib_resize_cq - Modifies the capacity of the CQ.
3405 * @cq: The CQ to resize.
3406 * @cqe: The minimum size of the CQ.
3408 * Users can examine the cq structure to determine the actual CQ size.
3410 int ib_resize_cq(struct ib_cq *cq, int cqe);
3413 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3414 * @cq: The CQ to modify.
3415 * @cq_count: number of CQEs that will trigger an event
3416 * @cq_period: max period of time in usec before triggering an event
3419 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3422 * ib_destroy_cq - Destroys the specified CQ.
3423 * @cq: The CQ to destroy.
3425 int ib_destroy_cq(struct ib_cq *cq);
3428 * ib_poll_cq - poll a CQ for completion(s)
3429 * @cq:the CQ being polled
3430 * @num_entries:maximum number of completions to return
3431 * @wc:array of at least @num_entries &struct ib_wc where completions
3434 * Poll a CQ for (possibly multiple) completions. If the return value
3435 * is < 0, an error occurred. If the return value is >= 0, it is the
3436 * number of completions returned. If the return value is
3437 * non-negative and < num_entries, then the CQ was emptied.
3439 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3442 return cq->device->poll_cq(cq, num_entries, wc);
3446 * ib_req_notify_cq - Request completion notification on a CQ.
3447 * @cq: The CQ to generate an event for.
3449 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3450 * to request an event on the next solicited event or next work
3451 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3452 * may also be |ed in to request a hint about missed events, as
3456 * < 0 means an error occurred while requesting notification
3457 * == 0 means notification was requested successfully, and if
3458 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3459 * were missed and it is safe to wait for another event. In
3460 * this case is it guaranteed that any work completions added
3461 * to the CQ since the last CQ poll will trigger a completion
3462 * notification event.
3463 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3464 * in. It means that the consumer must poll the CQ again to
3465 * make sure it is empty to avoid missing an event because of a
3466 * race between requesting notification and an entry being
3467 * added to the CQ. This return value means it is possible
3468 * (but not guaranteed) that a work completion has been added
3469 * to the CQ since the last poll without triggering a
3470 * completion notification event.
3472 static inline int ib_req_notify_cq(struct ib_cq *cq,
3473 enum ib_cq_notify_flags flags)
3475 return cq->device->req_notify_cq(cq, flags);
3479 * ib_req_ncomp_notif - Request completion notification when there are
3480 * at least the specified number of unreaped completions on the CQ.
3481 * @cq: The CQ to generate an event for.
3482 * @wc_cnt: The number of unreaped completions that should be on the
3483 * CQ before an event is generated.
3485 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3487 return cq->device->req_ncomp_notif ?
3488 cq->device->req_ncomp_notif(cq, wc_cnt) :
3493 * ib_dma_mapping_error - check a DMA addr for error
3494 * @dev: The device for which the dma_addr was created
3495 * @dma_addr: The DMA address to check
3497 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3499 return dma_mapping_error(dev->dma_device, dma_addr);
3503 * ib_dma_map_single - Map a kernel virtual address to DMA address
3504 * @dev: The device for which the dma_addr is to be created
3505 * @cpu_addr: The kernel virtual address
3506 * @size: The size of the region in bytes
3507 * @direction: The direction of the DMA
3509 static inline u64 ib_dma_map_single(struct ib_device *dev,
3510 void *cpu_addr, size_t size,
3511 enum dma_data_direction direction)
3513 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3517 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3518 * @dev: The device for which the DMA address was created
3519 * @addr: The DMA address
3520 * @size: The size of the region in bytes
3521 * @direction: The direction of the DMA
3523 static inline void ib_dma_unmap_single(struct ib_device *dev,
3524 u64 addr, size_t size,
3525 enum dma_data_direction direction)
3527 dma_unmap_single(dev->dma_device, addr, size, direction);
3531 * ib_dma_map_page - Map a physical page to DMA address
3532 * @dev: The device for which the dma_addr is to be created
3533 * @page: The page to be mapped
3534 * @offset: The offset within the page
3535 * @size: The size of the region in bytes
3536 * @direction: The direction of the DMA
3538 static inline u64 ib_dma_map_page(struct ib_device *dev,
3540 unsigned long offset,
3542 enum dma_data_direction direction)
3544 return dma_map_page(dev->dma_device, page, offset, size, direction);
3548 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3549 * @dev: The device for which the DMA address was created
3550 * @addr: The DMA address
3551 * @size: The size of the region in bytes
3552 * @direction: The direction of the DMA
3554 static inline void ib_dma_unmap_page(struct ib_device *dev,
3555 u64 addr, size_t size,
3556 enum dma_data_direction direction)
3558 dma_unmap_page(dev->dma_device, addr, size, direction);
3562 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3563 * @dev: The device for which the DMA addresses are to be created
3564 * @sg: The array of scatter/gather entries
3565 * @nents: The number of scatter/gather entries
3566 * @direction: The direction of the DMA
3568 static inline int ib_dma_map_sg(struct ib_device *dev,
3569 struct scatterlist *sg, int nents,
3570 enum dma_data_direction direction)
3572 return dma_map_sg(dev->dma_device, sg, nents, direction);
3576 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3577 * @dev: The device for which the DMA addresses were created
3578 * @sg: The array of scatter/gather entries
3579 * @nents: The number of scatter/gather entries
3580 * @direction: The direction of the DMA
3582 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3583 struct scatterlist *sg, int nents,
3584 enum dma_data_direction direction)
3586 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3589 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3590 struct scatterlist *sg, int nents,
3591 enum dma_data_direction direction,
3592 unsigned long dma_attrs)
3594 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3598 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3599 struct scatterlist *sg, int nents,
3600 enum dma_data_direction direction,
3601 unsigned long dma_attrs)
3603 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3606 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3607 * @dev: The device for which the DMA addresses were created
3608 * @sg: The scatter/gather entry
3610 * Note: this function is obsolete. To do: change all occurrences of
3611 * ib_sg_dma_address() into sg_dma_address().
3613 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3614 struct scatterlist *sg)
3616 return sg_dma_address(sg);
3620 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3621 * @dev: The device for which the DMA addresses were created
3622 * @sg: The scatter/gather entry
3624 * Note: this function is obsolete. To do: change all occurrences of
3625 * ib_sg_dma_len() into sg_dma_len().
3627 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3628 struct scatterlist *sg)
3630 return sg_dma_len(sg);
3634 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3635 * @dev: The device for which the DMA address was created
3636 * @addr: The DMA address
3637 * @size: The size of the region in bytes
3638 * @dir: The direction of the DMA
3640 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3643 enum dma_data_direction dir)
3645 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3649 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3650 * @dev: The device for which the DMA address was created
3651 * @addr: The DMA address
3652 * @size: The size of the region in bytes
3653 * @dir: The direction of the DMA
3655 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3658 enum dma_data_direction dir)
3660 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3664 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3665 * @dev: The device for which the DMA address is requested
3666 * @size: The size of the region to allocate in bytes
3667 * @dma_handle: A pointer for returning the DMA address of the region
3668 * @flag: memory allocator flags
3670 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3672 dma_addr_t *dma_handle,
3675 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3679 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3680 * @dev: The device for which the DMA addresses were allocated
3681 * @size: The size of the region
3682 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3683 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3685 static inline void ib_dma_free_coherent(struct ib_device *dev,
3686 size_t size, void *cpu_addr,
3687 dma_addr_t dma_handle)
3689 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3693 * ib_dereg_mr - Deregisters a memory region and removes it from the
3694 * HCA translation table.
3695 * @mr: The memory region to deregister.
3697 * This function can fail, if the memory region has memory windows bound to it.
3699 int ib_dereg_mr(struct ib_mr *mr);
3701 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3702 enum ib_mr_type mr_type,
3706 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3708 * @mr - struct ib_mr pointer to be updated.
3709 * @newkey - new key to be used.
3711 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3713 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3714 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3718 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3719 * for calculating a new rkey for type 2 memory windows.
3720 * @rkey - the rkey to increment.
3722 static inline u32 ib_inc_rkey(u32 rkey)
3724 const u32 mask = 0x000000ff;
3725 return ((rkey + 1) & mask) | (rkey & ~mask);
3729 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3730 * @pd: The protection domain associated with the unmapped region.
3731 * @mr_access_flags: Specifies the memory access rights.
3732 * @fmr_attr: Attributes of the unmapped region.
3734 * A fast memory region must be mapped before it can be used as part of
3737 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3738 int mr_access_flags,
3739 struct ib_fmr_attr *fmr_attr);
3742 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3743 * @fmr: The fast memory region to associate with the pages.
3744 * @page_list: An array of physical pages to map to the fast memory region.
3745 * @list_len: The number of pages in page_list.
3746 * @iova: The I/O virtual address to use with the mapped region.
3748 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3749 u64 *page_list, int list_len,
3752 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3756 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3757 * @fmr_list: A linked list of fast memory regions to unmap.
3759 int ib_unmap_fmr(struct list_head *fmr_list);
3762 * ib_dealloc_fmr - Deallocates a fast memory region.
3763 * @fmr: The fast memory region to deallocate.
3765 int ib_dealloc_fmr(struct ib_fmr *fmr);
3768 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3769 * @qp: QP to attach to the multicast group. The QP must be type
3771 * @gid: Multicast group GID.
3772 * @lid: Multicast group LID in host byte order.
3774 * In order to send and receive multicast packets, subnet
3775 * administration must have created the multicast group and configured
3776 * the fabric appropriately. The port associated with the specified
3777 * QP must also be a member of the multicast group.
3779 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3782 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3783 * @qp: QP to detach from the multicast group.
3784 * @gid: Multicast group GID.
3785 * @lid: Multicast group LID in host byte order.
3787 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3790 * ib_alloc_xrcd - Allocates an XRC domain.
3791 * @device: The device on which to allocate the XRC domain.
3792 * @caller: Module name for kernel consumers
3794 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3795 #define ib_alloc_xrcd(device) \
3796 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3799 * ib_dealloc_xrcd - Deallocates an XRC domain.
3800 * @xrcd: The XRC domain to deallocate.
3802 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3804 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3805 struct ib_flow_attr *flow_attr, int domain);
3806 int ib_destroy_flow(struct ib_flow *flow_id);
3808 static inline int ib_check_mr_access(int flags)
3811 * Local write permission is required if remote write or
3812 * remote atomic permission is also requested.
3814 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3815 !(flags & IB_ACCESS_LOCAL_WRITE))
3821 static inline bool ib_access_writable(int access_flags)
3824 * We have writable memory backing the MR if any of the following
3825 * access flags are set. "Local write" and "remote write" obviously
3826 * require write access. "Remote atomic" can do things like fetch and
3827 * add, which will modify memory, and "MW bind" can change permissions
3828 * by binding a window.
3830 return access_flags &
3831 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3832 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3836 * ib_check_mr_status: lightweight check of MR status.
3837 * This routine may provide status checks on a selected
3838 * ib_mr. first use is for signature status check.
3840 * @mr: A memory region.
3841 * @check_mask: Bitmask of which checks to perform from
3842 * ib_mr_status_check enumeration.
3843 * @mr_status: The container of relevant status checks.
3844 * failed checks will be indicated in the status bitmask
3845 * and the relevant info shall be in the error item.
3847 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3848 struct ib_mr_status *mr_status);
3850 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3851 u16 pkey, const union ib_gid *gid,
3852 const struct sockaddr *addr);
3853 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3854 struct ib_wq_init_attr *init_attr);
3855 int ib_destroy_wq(struct ib_wq *wq);
3856 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3858 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3859 struct ib_rwq_ind_table_init_attr*
3860 wq_ind_table_init_attr);
3861 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3863 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3864 unsigned int *sg_offset, unsigned int page_size);
3867 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3868 unsigned int *sg_offset, unsigned int page_size)
3872 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3878 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3879 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3881 void ib_drain_rq(struct ib_qp *qp);
3882 void ib_drain_sq(struct ib_qp *qp);
3883 void ib_drain_qp(struct ib_qp *qp);
3885 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3887 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3889 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3890 return attr->roce.dmac;
3894 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3896 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3897 attr->ib.dlid = (u16)dlid;
3898 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3899 attr->opa.dlid = dlid;
3902 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3904 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3905 return attr->ib.dlid;
3906 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3907 return attr->opa.dlid;
3911 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3916 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3921 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3924 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3925 attr->ib.src_path_bits = src_path_bits;
3926 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3927 attr->opa.src_path_bits = src_path_bits;
3930 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3932 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3933 return attr->ib.src_path_bits;
3934 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3935 return attr->opa.src_path_bits;
3939 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3942 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3943 attr->opa.make_grd = make_grd;
3946 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3948 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3949 return attr->opa.make_grd;
3953 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3955 attr->port_num = port_num;
3958 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3960 return attr->port_num;
3963 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3966 attr->static_rate = static_rate;
3969 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3971 return attr->static_rate;
3974 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3975 enum ib_ah_flags flag)
3977 attr->ah_flags = flag;
3980 static inline enum ib_ah_flags
3981 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3983 return attr->ah_flags;
3986 static inline const struct ib_global_route
3987 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3992 /*To retrieve and modify the grh */
3993 static inline struct ib_global_route
3994 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3999 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4001 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4003 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4006 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4009 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4011 grh->dgid.global.subnet_prefix = prefix;
4014 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4017 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4019 grh->dgid.global.interface_id = if_id;
4022 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4023 union ib_gid *dgid, u32 flow_label,
4024 u8 sgid_index, u8 hop_limit,
4027 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4029 attr->ah_flags = IB_AH_GRH;
4032 grh->flow_label = flow_label;
4033 grh->sgid_index = sgid_index;
4034 grh->hop_limit = hop_limit;
4035 grh->traffic_class = traffic_class;
4039 * rdma_ah_find_type - Return address handle type.
4041 * @dev: Device to be checked
4042 * @port_num: Port number
4044 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4047 if (rdma_protocol_roce(dev, port_num))
4048 return RDMA_AH_ATTR_TYPE_ROCE;
4049 if (rdma_protocol_ib(dev, port_num)) {
4050 if (rdma_cap_opa_ah(dev, port_num))
4051 return RDMA_AH_ATTR_TYPE_OPA;
4052 return RDMA_AH_ATTR_TYPE_IB;
4055 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4059 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4060 * In the current implementation the only way to get
4061 * get the 32bit lid is from other sources for OPA.
4062 * For IB, lids will always be 16bits so cast the
4063 * value accordingly.
4067 static inline u16 ib_lid_cpu16(u32 lid)
4069 WARN_ON_ONCE(lid & 0xFFFF0000);
4074 * ib_lid_be16 - Return lid in 16bit BE encoding.
4078 static inline __be16 ib_lid_be16(u32 lid)
4080 WARN_ON_ONCE(lid & 0xFFFF0000);
4081 return cpu_to_be16((u16)lid);
4085 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4087 * @device: the rdma device
4088 * @comp_vector: index of completion vector
4090 * Returns NULL on failure, otherwise a corresponding cpu map of the
4091 * completion vector (returns all-cpus map if the device driver doesn't
4092 * implement get_vector_affinity).
4094 static inline const struct cpumask *
4095 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4097 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4098 !device->get_vector_affinity)
4101 return device->get_vector_affinity(device, comp_vector);
4106 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4107 * and add their gids, as needed, to the relevant RoCE devices.
4109 * @device: the rdma device
4111 void rdma_roce_rescan_device(struct ib_device *ibdev);
4113 #endif /* IB_VERBS_H */