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>
67 extern struct workqueue_struct
*ib_wq
;
68 extern struct workqueue_struct
*ib_comp_wq
;
78 extern union ib_gid zgid
;
81 /* If link layer is Ethernet, this is RoCE V1 */
84 IB_GID_TYPE_ROCE_UDP_ENCAP
= 1,
88 #define ROCE_V2_UDP_DPORT 4791
90 enum ib_gid_type gid_type
;
91 struct net_device
*ndev
;
95 /* IB values map to NodeInfo:NodeType. */
105 /* set the local administered indication */
106 IB_SA_WELL_KNOWN_GUID
= BIT_ULL(57) | 2,
109 enum rdma_transport_type
{
111 RDMA_TRANSPORT_IWARP
,
112 RDMA_TRANSPORT_USNIC
,
113 RDMA_TRANSPORT_USNIC_UDP
116 enum rdma_protocol_type
{
120 RDMA_PROTOCOL_USNIC_UDP
123 __attribute_const__
enum rdma_transport_type
124 rdma_node_get_transport(enum rdma_node_type node_type
);
126 enum rdma_network_type
{
128 RDMA_NETWORK_ROCE_V1
= RDMA_NETWORK_IB
,
133 static inline enum ib_gid_type
ib_network_to_gid_type(enum rdma_network_type network_type
)
135 if (network_type
== RDMA_NETWORK_IPV4
||
136 network_type
== RDMA_NETWORK_IPV6
)
137 return IB_GID_TYPE_ROCE_UDP_ENCAP
;
139 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
140 return IB_GID_TYPE_IB
;
143 static inline enum rdma_network_type
ib_gid_to_network_type(enum ib_gid_type gid_type
,
146 if (gid_type
== IB_GID_TYPE_IB
)
147 return RDMA_NETWORK_IB
;
149 if (ipv6_addr_v4mapped((struct in6_addr
*)gid
))
150 return RDMA_NETWORK_IPV4
;
152 return RDMA_NETWORK_IPV6
;
155 enum rdma_link_layer
{
156 IB_LINK_LAYER_UNSPECIFIED
,
157 IB_LINK_LAYER_INFINIBAND
,
158 IB_LINK_LAYER_ETHERNET
,
161 enum ib_device_cap_flags
{
162 IB_DEVICE_RESIZE_MAX_WR
= (1 << 0),
163 IB_DEVICE_BAD_PKEY_CNTR
= (1 << 1),
164 IB_DEVICE_BAD_QKEY_CNTR
= (1 << 2),
165 IB_DEVICE_RAW_MULTI
= (1 << 3),
166 IB_DEVICE_AUTO_PATH_MIG
= (1 << 4),
167 IB_DEVICE_CHANGE_PHY_PORT
= (1 << 5),
168 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1 << 6),
169 IB_DEVICE_CURR_QP_STATE_MOD
= (1 << 7),
170 IB_DEVICE_SHUTDOWN_PORT
= (1 << 8),
171 IB_DEVICE_INIT_TYPE
= (1 << 9),
172 IB_DEVICE_PORT_ACTIVE_EVENT
= (1 << 10),
173 IB_DEVICE_SYS_IMAGE_GUID
= (1 << 11),
174 IB_DEVICE_RC_RNR_NAK_GEN
= (1 << 12),
175 IB_DEVICE_SRQ_RESIZE
= (1 << 13),
176 IB_DEVICE_N_NOTIFY_CQ
= (1 << 14),
179 * This device supports a per-device lkey or stag that can be
180 * used without performing a memory registration for the local
181 * memory. Note that ULPs should never check this flag, but
182 * instead of use the local_dma_lkey flag in the ib_pd structure,
183 * which will always contain a usable lkey.
185 IB_DEVICE_LOCAL_DMA_LKEY
= (1 << 15),
186 IB_DEVICE_RESERVED
/* old SEND_W_INV */ = (1 << 16),
187 IB_DEVICE_MEM_WINDOW
= (1 << 17),
189 * Devices should set IB_DEVICE_UD_IP_SUM if they support
190 * insertion of UDP and TCP checksum on outgoing UD IPoIB
191 * messages and can verify the validity of checksum for
192 * incoming messages. Setting this flag implies that the
193 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
195 IB_DEVICE_UD_IP_CSUM
= (1 << 18),
196 IB_DEVICE_UD_TSO
= (1 << 19),
197 IB_DEVICE_XRC
= (1 << 20),
200 * This device supports the IB "base memory management extension",
201 * which includes support for fast registrations (IB_WR_REG_MR,
202 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
203 * also be set by any iWarp device which must support FRs to comply
204 * to the iWarp verbs spec. iWarp devices also support the
205 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
208 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1 << 21),
209 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1 << 22),
210 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1 << 23),
211 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1 << 24),
212 IB_DEVICE_RC_IP_CSUM
= (1 << 25),
213 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
214 IB_DEVICE_RAW_IP_CSUM
= (1 << 26),
216 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
217 * support execution of WQEs that involve synchronization
218 * of I/O operations with single completion queue managed
221 IB_DEVICE_CROSS_CHANNEL
= (1 << 27),
222 IB_DEVICE_MANAGED_FLOW_STEERING
= (1 << 29),
223 IB_DEVICE_SIGNATURE_HANDOVER
= (1 << 30),
224 IB_DEVICE_ON_DEMAND_PAGING
= (1ULL << 31),
225 IB_DEVICE_SG_GAPS_REG
= (1ULL << 32),
226 IB_DEVICE_VIRTUAL_FUNCTION
= (1ULL << 33),
227 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
228 IB_DEVICE_RAW_SCATTER_FCS
= (1ULL << 34),
229 IB_DEVICE_RDMA_NETDEV_OPA_VNIC
= (1ULL << 35),
232 enum ib_signature_prot_cap
{
233 IB_PROT_T10DIF_TYPE_1
= 1,
234 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
235 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
238 enum ib_signature_guard_cap
{
239 IB_GUARD_T10DIF_CRC
= 1,
240 IB_GUARD_T10DIF_CSUM
= 1 << 1,
249 enum ib_odp_general_cap_bits
{
250 IB_ODP_SUPPORT
= 1 << 0,
251 IB_ODP_SUPPORT_IMPLICIT
= 1 << 1,
254 enum ib_odp_transport_cap_bits
{
255 IB_ODP_SUPPORT_SEND
= 1 << 0,
256 IB_ODP_SUPPORT_RECV
= 1 << 1,
257 IB_ODP_SUPPORT_WRITE
= 1 << 2,
258 IB_ODP_SUPPORT_READ
= 1 << 3,
259 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
263 uint64_t general_caps
;
265 uint32_t rc_odp_caps
;
266 uint32_t uc_odp_caps
;
267 uint32_t ud_odp_caps
;
268 } per_transport_caps
;
272 /* Corresponding bit will be set if qp type from
273 * 'enum ib_qp_type' is supported, e.g.
274 * supported_qpts |= 1 << IB_QPT_UD
277 u32 max_rwq_indirection_tables
;
278 u32 max_rwq_indirection_table_size
;
281 enum ib_cq_creation_flags
{
282 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
283 IB_CQ_FLAGS_IGNORE_OVERRUN
= 1 << 1,
286 struct ib_cq_init_attr
{
292 struct ib_device_attr
{
294 __be64 sys_image_guid
;
302 u64 device_cap_flags
;
312 int max_qp_init_rd_atom
;
313 int max_ee_init_rd_atom
;
314 enum ib_atomic_cap atomic_cap
;
315 enum ib_atomic_cap masked_atomic_cap
;
322 int max_mcast_qp_attach
;
323 int max_total_mcast_qp_attach
;
330 unsigned int max_fast_reg_page_list_len
;
332 u8 local_ca_ack_delay
;
335 struct ib_odp_caps odp_caps
;
336 uint64_t timestamp_mask
;
337 uint64_t hca_core_clock
; /* in KHZ */
338 struct ib_rss_caps rss_caps
;
340 u32 raw_packet_caps
; /* Use ib_raw_packet_caps enum */
351 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
354 case IB_MTU_256
: return 256;
355 case IB_MTU_512
: return 512;
356 case IB_MTU_1024
: return 1024;
357 case IB_MTU_2048
: return 2048;
358 case IB_MTU_4096
: return 4096;
363 static inline enum ib_mtu
ib_mtu_int_to_enum(int mtu
)
367 else if (mtu
>= 2048)
369 else if (mtu
>= 1024)
383 IB_PORT_ACTIVE_DEFER
= 5
386 enum ib_port_cap_flags
{
388 IB_PORT_NOTICE_SUP
= 1 << 2,
389 IB_PORT_TRAP_SUP
= 1 << 3,
390 IB_PORT_OPT_IPD_SUP
= 1 << 4,
391 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
392 IB_PORT_SL_MAP_SUP
= 1 << 6,
393 IB_PORT_MKEY_NVRAM
= 1 << 7,
394 IB_PORT_PKEY_NVRAM
= 1 << 8,
395 IB_PORT_LED_INFO_SUP
= 1 << 9,
396 IB_PORT_SM_DISABLED
= 1 << 10,
397 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
398 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
399 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
400 IB_PORT_CM_SUP
= 1 << 16,
401 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
402 IB_PORT_REINIT_SUP
= 1 << 18,
403 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
404 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
405 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
406 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
407 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
408 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
409 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
410 IB_PORT_IP_BASED_GIDS
= 1 << 26,
420 static inline int ib_width_enum_to_int(enum ib_port_width width
)
423 case IB_WIDTH_1X
: return 1;
424 case IB_WIDTH_4X
: return 4;
425 case IB_WIDTH_8X
: return 8;
426 case IB_WIDTH_12X
: return 12;
442 * struct rdma_hw_stats
443 * @timestamp - Used by the core code to track when the last update was
444 * @lifespan - Used by the core code to determine how old the counters
445 * should be before being updated again. Stored in jiffies, defaults
446 * to 10 milliseconds, drivers can override the default be specifying
447 * their own value during their allocation routine.
448 * @name - Array of pointers to static names used for the counters in
450 * @num_counters - How many hardware counters there are. If name is
451 * shorter than this number, a kernel oops will result. Driver authors
452 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
453 * in their code to prevent this.
454 * @value - Array of u64 counters that are accessed by the sysfs code and
455 * filled in by the drivers get_stats routine
457 struct rdma_hw_stats
{
458 unsigned long timestamp
;
459 unsigned long lifespan
;
460 const char * const *names
;
465 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
467 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
469 * @names - Array of static const char *
470 * @num_counters - How many elements in array
471 * @lifespan - How many milliseconds between updates
473 static inline struct rdma_hw_stats
*rdma_alloc_hw_stats_struct(
474 const char * const *names
, int num_counters
,
475 unsigned long lifespan
)
477 struct rdma_hw_stats
*stats
;
479 stats
= kzalloc(sizeof(*stats
) + num_counters
* sizeof(u64
),
483 stats
->names
= names
;
484 stats
->num_counters
= num_counters
;
485 stats
->lifespan
= msecs_to_jiffies(lifespan
);
491 /* Define bits for the various functionality this port needs to be supported by
494 /* Management 0x00000FFF */
495 #define RDMA_CORE_CAP_IB_MAD 0x00000001
496 #define RDMA_CORE_CAP_IB_SMI 0x00000002
497 #define RDMA_CORE_CAP_IB_CM 0x00000004
498 #define RDMA_CORE_CAP_IW_CM 0x00000008
499 #define RDMA_CORE_CAP_IB_SA 0x00000010
500 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
502 /* Address format 0x000FF000 */
503 #define RDMA_CORE_CAP_AF_IB 0x00001000
504 #define RDMA_CORE_CAP_ETH_AH 0x00002000
505 #define RDMA_CORE_CAP_OPA_AH 0x00004000
507 /* Protocol 0xFFF00000 */
508 #define RDMA_CORE_CAP_PROT_IB 0x00100000
509 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
510 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
511 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
512 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
513 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
515 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
516 | RDMA_CORE_CAP_IB_MAD \
517 | RDMA_CORE_CAP_IB_SMI \
518 | RDMA_CORE_CAP_IB_CM \
519 | RDMA_CORE_CAP_IB_SA \
520 | RDMA_CORE_CAP_AF_IB)
521 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
522 | RDMA_CORE_CAP_IB_MAD \
523 | RDMA_CORE_CAP_IB_CM \
524 | RDMA_CORE_CAP_AF_IB \
525 | RDMA_CORE_CAP_ETH_AH)
526 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
527 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
528 | RDMA_CORE_CAP_IB_MAD \
529 | RDMA_CORE_CAP_IB_CM \
530 | RDMA_CORE_CAP_AF_IB \
531 | RDMA_CORE_CAP_ETH_AH)
532 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
533 | RDMA_CORE_CAP_IW_CM)
534 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
535 | RDMA_CORE_CAP_OPA_MAD)
537 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
539 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
541 struct ib_port_attr
{
543 enum ib_port_state state
;
545 enum ib_mtu active_mtu
;
565 enum ib_device_modify_flags
{
566 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
567 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
570 #define IB_DEVICE_NODE_DESC_MAX 64
572 struct ib_device_modify
{
574 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
577 enum ib_port_modify_flags
{
578 IB_PORT_SHUTDOWN
= 1,
579 IB_PORT_INIT_TYPE
= (1<<2),
580 IB_PORT_RESET_QKEY_CNTR
= (1<<3),
581 IB_PORT_OPA_MASK_CHG
= (1<<4)
584 struct ib_port_modify
{
585 u32 set_port_cap_mask
;
586 u32 clr_port_cap_mask
;
594 IB_EVENT_QP_ACCESS_ERR
,
598 IB_EVENT_PATH_MIG_ERR
,
599 IB_EVENT_DEVICE_FATAL
,
600 IB_EVENT_PORT_ACTIVE
,
603 IB_EVENT_PKEY_CHANGE
,
606 IB_EVENT_SRQ_LIMIT_REACHED
,
607 IB_EVENT_QP_LAST_WQE_REACHED
,
608 IB_EVENT_CLIENT_REREGISTER
,
613 const char *__attribute_const__
ib_event_msg(enum ib_event_type event
);
616 struct ib_device
*device
;
624 enum ib_event_type event
;
627 struct ib_event_handler
{
628 struct ib_device
*device
;
629 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
630 struct list_head list
;
633 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
635 (_ptr)->device = _device; \
636 (_ptr)->handler = _handler; \
637 INIT_LIST_HEAD(&(_ptr)->list); \
640 struct ib_global_route
{
649 __be32 version_tclass_flow
;
657 union rdma_network_hdr
{
660 /* The IB spec states that if it's IPv4, the header
661 * is located in the last 20 bytes of the header.
664 struct iphdr roce4grh
;
668 #define IB_QPN_MASK 0xFFFFFF
671 IB_MULTICAST_QPN
= 0xffffff
674 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
675 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
682 IB_RATE_PORT_CURRENT
= 0,
683 IB_RATE_2_5_GBPS
= 2,
691 IB_RATE_120_GBPS
= 10,
692 IB_RATE_14_GBPS
= 11,
693 IB_RATE_56_GBPS
= 12,
694 IB_RATE_112_GBPS
= 13,
695 IB_RATE_168_GBPS
= 14,
696 IB_RATE_25_GBPS
= 15,
697 IB_RATE_100_GBPS
= 16,
698 IB_RATE_200_GBPS
= 17,
699 IB_RATE_300_GBPS
= 18
703 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
704 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
705 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
706 * @rate: rate to convert.
708 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
711 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
712 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
713 * @rate: rate to convert.
715 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
719 * enum ib_mr_type - memory region type
720 * @IB_MR_TYPE_MEM_REG: memory region that is used for
721 * normal registration
722 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
723 * signature operations (data-integrity
725 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
726 * register any arbitrary sg lists (without
727 * the normal mr constraints - see
732 IB_MR_TYPE_SIGNATURE
,
738 * IB_SIG_TYPE_NONE: Unprotected.
739 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
741 enum ib_signature_type
{
747 * Signature T10-DIF block-guard types
748 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
749 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
751 enum ib_t10_dif_bg_type
{
757 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
759 * @bg_type: T10-DIF block guard type (CRC|CSUM)
760 * @pi_interval: protection information interval.
761 * @bg: seed of guard computation.
762 * @app_tag: application tag of guard block
763 * @ref_tag: initial guard block reference tag.
764 * @ref_remap: Indicate wethear the reftag increments each block
765 * @app_escape: Indicate to skip block check if apptag=0xffff
766 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
767 * @apptag_check_mask: check bitmask of application tag.
769 struct ib_t10_dif_domain
{
770 enum ib_t10_dif_bg_type bg_type
;
778 u16 apptag_check_mask
;
782 * struct ib_sig_domain - Parameters for signature domain
783 * @sig_type: specific signauture type
784 * @sig: union of all signature domain attributes that may
785 * be used to set domain layout.
787 struct ib_sig_domain
{
788 enum ib_signature_type sig_type
;
790 struct ib_t10_dif_domain dif
;
795 * struct ib_sig_attrs - Parameters for signature handover operation
796 * @check_mask: bitmask for signature byte check (8 bytes)
797 * @mem: memory domain layout desciptor.
798 * @wire: wire domain layout desciptor.
800 struct ib_sig_attrs
{
802 struct ib_sig_domain mem
;
803 struct ib_sig_domain wire
;
806 enum ib_sig_err_type
{
813 * struct ib_sig_err - signature error descriptor
816 enum ib_sig_err_type err_type
;
823 enum ib_mr_status_check
{
824 IB_MR_CHECK_SIG_STATUS
= 1,
828 * struct ib_mr_status - Memory region status container
830 * @fail_status: Bitmask of MR checks status. For each
831 * failed check a corresponding status bit is set.
832 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
835 struct ib_mr_status
{
837 struct ib_sig_err sig_err
;
841 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
843 * @mult: multiple to convert.
845 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
847 enum rdma_ah_attr_type
{
848 RDMA_AH_ATTR_TYPE_IB
,
849 RDMA_AH_ATTR_TYPE_ROCE
,
850 RDMA_AH_ATTR_TYPE_OPA
,
858 struct roce_ah_attr
{
867 struct rdma_ah_attr
{
868 struct ib_global_route grh
;
873 enum rdma_ah_attr_type type
;
875 struct ib_ah_attr ib
;
876 struct roce_ah_attr roce
;
877 struct opa_ah_attr opa
;
885 IB_WC_LOC_EEC_OP_ERR
,
890 IB_WC_LOC_ACCESS_ERR
,
891 IB_WC_REM_INV_REQ_ERR
,
892 IB_WC_REM_ACCESS_ERR
,
895 IB_WC_RNR_RETRY_EXC_ERR
,
896 IB_WC_LOC_RDD_VIOL_ERR
,
897 IB_WC_REM_INV_RD_REQ_ERR
,
900 IB_WC_INV_EEC_STATE_ERR
,
902 IB_WC_RESP_TIMEOUT_ERR
,
906 const char *__attribute_const__
ib_wc_status_msg(enum ib_wc_status status
);
917 IB_WC_MASKED_COMP_SWAP
,
918 IB_WC_MASKED_FETCH_ADD
,
920 * Set value of IB_WC_RECV so consumers can test if a completion is a
921 * receive by testing (opcode & IB_WC_RECV).
924 IB_WC_RECV_RDMA_WITH_IMM
929 IB_WC_WITH_IMM
= (1<<1),
930 IB_WC_WITH_INVALIDATE
= (1<<2),
931 IB_WC_IP_CSUM_OK
= (1<<3),
932 IB_WC_WITH_SMAC
= (1<<4),
933 IB_WC_WITH_VLAN
= (1<<5),
934 IB_WC_WITH_NETWORK_HDR_TYPE
= (1<<6),
940 struct ib_cqe
*wr_cqe
;
942 enum ib_wc_status status
;
943 enum ib_wc_opcode opcode
;
957 u8 port_num
; /* valid only for DR SMPs on switches */
963 enum ib_cq_notify_flags
{
964 IB_CQ_SOLICITED
= 1 << 0,
965 IB_CQ_NEXT_COMP
= 1 << 1,
966 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
967 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
975 enum ib_srq_attr_mask
{
976 IB_SRQ_MAX_WR
= 1 << 0,
977 IB_SRQ_LIMIT
= 1 << 1,
986 struct ib_srq_init_attr
{
987 void (*event_handler
)(struct ib_event
*, void *);
989 struct ib_srq_attr attr
;
990 enum ib_srq_type srq_type
;
994 struct ib_xrcd
*xrcd
;
1005 u32 max_inline_data
;
1008 * Maximum number of rdma_rw_ctx structures in flight at a time.
1009 * ib_create_qp() will calculate the right amount of neededed WRs
1010 * and MRs based on this.
1022 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1023 * here (and in that order) since the MAD layer uses them as
1024 * indices into a 2-entry table.
1033 IB_QPT_RAW_ETHERTYPE
,
1034 IB_QPT_RAW_PACKET
= 8,
1038 /* Reserve a range for qp types internal to the low level driver.
1039 * These qp types will not be visible at the IB core layer, so the
1040 * IB_QPT_MAX usages should not be affected in the core layer
1042 IB_QPT_RESERVED1
= 0x1000,
1054 enum ib_qp_create_flags
{
1055 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
1056 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
1057 IB_QP_CREATE_CROSS_CHANNEL
= 1 << 2,
1058 IB_QP_CREATE_MANAGED_SEND
= 1 << 3,
1059 IB_QP_CREATE_MANAGED_RECV
= 1 << 4,
1060 IB_QP_CREATE_NETIF_QP
= 1 << 5,
1061 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
1062 /* FREE = 1 << 7, */
1063 IB_QP_CREATE_SCATTER_FCS
= 1 << 8,
1064 IB_QP_CREATE_CVLAN_STRIPPING
= 1 << 9,
1065 IB_QP_CREATE_SOURCE_QPN
= 1 << 10,
1066 /* reserve bits 26-31 for low level drivers' internal use */
1067 IB_QP_CREATE_RESERVED_START
= 1 << 26,
1068 IB_QP_CREATE_RESERVED_END
= 1 << 31,
1072 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1073 * callback to destroy the passed in QP.
1076 struct ib_qp_init_attr
{
1077 void (*event_handler
)(struct ib_event
*, void *);
1079 struct ib_cq
*send_cq
;
1080 struct ib_cq
*recv_cq
;
1082 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1083 struct ib_qp_cap cap
;
1084 enum ib_sig_type sq_sig_type
;
1085 enum ib_qp_type qp_type
;
1086 enum ib_qp_create_flags create_flags
;
1089 * Only needed for special QP types, or when using the RW API.
1092 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1096 struct ib_qp_open_attr
{
1097 void (*event_handler
)(struct ib_event
*, void *);
1100 enum ib_qp_type qp_type
;
1103 enum ib_rnr_timeout
{
1104 IB_RNR_TIMER_655_36
= 0,
1105 IB_RNR_TIMER_000_01
= 1,
1106 IB_RNR_TIMER_000_02
= 2,
1107 IB_RNR_TIMER_000_03
= 3,
1108 IB_RNR_TIMER_000_04
= 4,
1109 IB_RNR_TIMER_000_06
= 5,
1110 IB_RNR_TIMER_000_08
= 6,
1111 IB_RNR_TIMER_000_12
= 7,
1112 IB_RNR_TIMER_000_16
= 8,
1113 IB_RNR_TIMER_000_24
= 9,
1114 IB_RNR_TIMER_000_32
= 10,
1115 IB_RNR_TIMER_000_48
= 11,
1116 IB_RNR_TIMER_000_64
= 12,
1117 IB_RNR_TIMER_000_96
= 13,
1118 IB_RNR_TIMER_001_28
= 14,
1119 IB_RNR_TIMER_001_92
= 15,
1120 IB_RNR_TIMER_002_56
= 16,
1121 IB_RNR_TIMER_003_84
= 17,
1122 IB_RNR_TIMER_005_12
= 18,
1123 IB_RNR_TIMER_007_68
= 19,
1124 IB_RNR_TIMER_010_24
= 20,
1125 IB_RNR_TIMER_015_36
= 21,
1126 IB_RNR_TIMER_020_48
= 22,
1127 IB_RNR_TIMER_030_72
= 23,
1128 IB_RNR_TIMER_040_96
= 24,
1129 IB_RNR_TIMER_061_44
= 25,
1130 IB_RNR_TIMER_081_92
= 26,
1131 IB_RNR_TIMER_122_88
= 27,
1132 IB_RNR_TIMER_163_84
= 28,
1133 IB_RNR_TIMER_245_76
= 29,
1134 IB_RNR_TIMER_327_68
= 30,
1135 IB_RNR_TIMER_491_52
= 31
1138 enum ib_qp_attr_mask
{
1140 IB_QP_CUR_STATE
= (1<<1),
1141 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
1142 IB_QP_ACCESS_FLAGS
= (1<<3),
1143 IB_QP_PKEY_INDEX
= (1<<4),
1144 IB_QP_PORT
= (1<<5),
1145 IB_QP_QKEY
= (1<<6),
1147 IB_QP_PATH_MTU
= (1<<8),
1148 IB_QP_TIMEOUT
= (1<<9),
1149 IB_QP_RETRY_CNT
= (1<<10),
1150 IB_QP_RNR_RETRY
= (1<<11),
1151 IB_QP_RQ_PSN
= (1<<12),
1152 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
1153 IB_QP_ALT_PATH
= (1<<14),
1154 IB_QP_MIN_RNR_TIMER
= (1<<15),
1155 IB_QP_SQ_PSN
= (1<<16),
1156 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
1157 IB_QP_PATH_MIG_STATE
= (1<<18),
1158 IB_QP_CAP
= (1<<19),
1159 IB_QP_DEST_QPN
= (1<<20),
1160 IB_QP_RESERVED1
= (1<<21),
1161 IB_QP_RESERVED2
= (1<<22),
1162 IB_QP_RESERVED3
= (1<<23),
1163 IB_QP_RESERVED4
= (1<<24),
1164 IB_QP_RATE_LIMIT
= (1<<25),
1189 enum ib_qp_state qp_state
;
1190 enum ib_qp_state cur_qp_state
;
1191 enum ib_mtu path_mtu
;
1192 enum ib_mig_state path_mig_state
;
1197 int qp_access_flags
;
1198 struct ib_qp_cap cap
;
1199 struct rdma_ah_attr ah_attr
;
1200 struct rdma_ah_attr alt_ah_attr
;
1203 u8 en_sqd_async_notify
;
1206 u8 max_dest_rd_atomic
;
1219 IB_WR_RDMA_WRITE_WITH_IMM
,
1221 IB_WR_SEND_WITH_IMM
,
1223 IB_WR_ATOMIC_CMP_AND_SWP
,
1224 IB_WR_ATOMIC_FETCH_AND_ADD
,
1226 IB_WR_SEND_WITH_INV
,
1227 IB_WR_RDMA_READ_WITH_INV
,
1230 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1231 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1233 /* reserve values for low level drivers' internal use.
1234 * These values will not be used at all in the ib core layer.
1236 IB_WR_RESERVED1
= 0xf0,
1248 enum ib_send_flags
{
1250 IB_SEND_SIGNALED
= (1<<1),
1251 IB_SEND_SOLICITED
= (1<<2),
1252 IB_SEND_INLINE
= (1<<3),
1253 IB_SEND_IP_CSUM
= (1<<4),
1255 /* reserve bits 26-31 for low level drivers' internal use */
1256 IB_SEND_RESERVED_START
= (1 << 26),
1257 IB_SEND_RESERVED_END
= (1 << 31),
1267 void (*done
)(struct ib_cq
*cq
, struct ib_wc
*wc
);
1271 struct ib_send_wr
*next
;
1274 struct ib_cqe
*wr_cqe
;
1276 struct ib_sge
*sg_list
;
1278 enum ib_wr_opcode opcode
;
1282 u32 invalidate_rkey
;
1287 struct ib_send_wr wr
;
1292 static inline struct ib_rdma_wr
*rdma_wr(struct ib_send_wr
*wr
)
1294 return container_of(wr
, struct ib_rdma_wr
, wr
);
1297 struct ib_atomic_wr
{
1298 struct ib_send_wr wr
;
1302 u64 compare_add_mask
;
1307 static inline struct ib_atomic_wr
*atomic_wr(struct ib_send_wr
*wr
)
1309 return container_of(wr
, struct ib_atomic_wr
, wr
);
1313 struct ib_send_wr wr
;
1320 u16 pkey_index
; /* valid for GSI only */
1321 u8 port_num
; /* valid for DR SMPs on switch only */
1324 static inline struct ib_ud_wr
*ud_wr(struct ib_send_wr
*wr
)
1326 return container_of(wr
, struct ib_ud_wr
, wr
);
1330 struct ib_send_wr wr
;
1336 static inline struct ib_reg_wr
*reg_wr(struct ib_send_wr
*wr
)
1338 return container_of(wr
, struct ib_reg_wr
, wr
);
1341 struct ib_sig_handover_wr
{
1342 struct ib_send_wr wr
;
1343 struct ib_sig_attrs
*sig_attrs
;
1344 struct ib_mr
*sig_mr
;
1346 struct ib_sge
*prot
;
1349 static inline struct ib_sig_handover_wr
*sig_handover_wr(struct ib_send_wr
*wr
)
1351 return container_of(wr
, struct ib_sig_handover_wr
, wr
);
1355 struct ib_recv_wr
*next
;
1358 struct ib_cqe
*wr_cqe
;
1360 struct ib_sge
*sg_list
;
1364 enum ib_access_flags
{
1365 IB_ACCESS_LOCAL_WRITE
= 1,
1366 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1367 IB_ACCESS_REMOTE_READ
= (1<<2),
1368 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1369 IB_ACCESS_MW_BIND
= (1<<4),
1370 IB_ZERO_BASED
= (1<<5),
1371 IB_ACCESS_ON_DEMAND
= (1<<6),
1372 IB_ACCESS_HUGETLB
= (1<<7),
1376 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1377 * are hidden here instead of a uapi header!
1379 enum ib_mr_rereg_flags
{
1380 IB_MR_REREG_TRANS
= 1,
1381 IB_MR_REREG_PD
= (1<<1),
1382 IB_MR_REREG_ACCESS
= (1<<2),
1383 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1386 struct ib_fmr_attr
{
1394 enum rdma_remove_reason
{
1395 /* Userspace requested uobject deletion. Call could fail */
1396 RDMA_REMOVE_DESTROY
,
1397 /* Context deletion. This call should delete the actual object itself */
1399 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1400 RDMA_REMOVE_DRIVER_REMOVE
,
1401 /* Context is being cleaned-up, but commit was just completed */
1402 RDMA_REMOVE_DURING_CLEANUP
,
1405 struct ib_rdmacg_object
{
1406 #ifdef CONFIG_CGROUP_RDMA
1407 struct rdma_cgroup
*cg
; /* owner rdma cgroup */
1411 struct ib_ucontext
{
1412 struct ib_device
*device
;
1413 struct ib_uverbs_file
*ufile
;
1416 /* locking the uobjects_list */
1417 struct mutex uobjects_lock
;
1418 struct list_head uobjects
;
1419 /* protects cleanup process from other actions */
1420 struct rw_semaphore cleanup_rwsem
;
1421 enum rdma_remove_reason cleanup_reason
;
1424 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1425 struct rb_root umem_tree
;
1427 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1428 * mmu notifiers registration.
1430 struct rw_semaphore umem_rwsem
;
1431 void (*invalidate_range
)(struct ib_umem
*umem
,
1432 unsigned long start
, unsigned long end
);
1434 struct mmu_notifier mn
;
1435 atomic_t notifier_count
;
1436 /* A list of umems that don't have private mmu notifier counters yet. */
1437 struct list_head no_private_counters
;
1441 struct ib_rdmacg_object cg_obj
;
1445 u64 user_handle
; /* handle given to us by userspace */
1446 struct ib_ucontext
*context
; /* associated user context */
1447 void *object
; /* containing object */
1448 struct list_head list
; /* link to context's list */
1449 struct ib_rdmacg_object cg_obj
; /* rdmacg object */
1450 int id
; /* index into kernel idr */
1452 atomic_t usecnt
; /* protects exclusive access */
1453 struct rcu_head rcu
; /* kfree_rcu() overhead */
1455 const struct uverbs_obj_type
*type
;
1458 struct ib_uobject_file
{
1459 struct ib_uobject uobj
;
1460 /* ufile contains the lock between context release and file close */
1461 struct ib_uverbs_file
*ufile
;
1465 const void __user
*inbuf
;
1466 void __user
*outbuf
;
1474 struct ib_device
*device
;
1475 struct ib_uobject
*uobject
;
1476 atomic_t usecnt
; /* count all resources */
1478 u32 unsafe_global_rkey
;
1481 * Implementation details of the RDMA core, don't use in drivers:
1483 struct ib_mr
*__internal_mr
;
1487 struct ib_device
*device
;
1488 atomic_t usecnt
; /* count all exposed resources */
1489 struct inode
*inode
;
1491 struct mutex tgt_qp_mutex
;
1492 struct list_head tgt_qp_list
;
1496 struct ib_device
*device
;
1498 struct ib_uobject
*uobject
;
1499 enum rdma_ah_attr_type type
;
1502 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1504 enum ib_poll_context
{
1505 IB_POLL_DIRECT
, /* caller context, no hw completions */
1506 IB_POLL_SOFTIRQ
, /* poll from softirq context */
1507 IB_POLL_WORKQUEUE
, /* poll from workqueue */
1511 struct ib_device
*device
;
1512 struct ib_uobject
*uobject
;
1513 ib_comp_handler comp_handler
;
1514 void (*event_handler
)(struct ib_event
*, void *);
1517 atomic_t usecnt
; /* count number of work queues */
1518 enum ib_poll_context poll_ctx
;
1521 struct irq_poll iop
;
1522 struct work_struct work
;
1527 struct ib_device
*device
;
1529 struct ib_uobject
*uobject
;
1530 void (*event_handler
)(struct ib_event
*, void *);
1532 enum ib_srq_type srq_type
;
1537 struct ib_xrcd
*xrcd
;
1544 enum ib_raw_packet_caps
{
1545 /* Strip cvlan from incoming packet and report it in the matching work
1546 * completion is supported.
1548 IB_RAW_PACKET_CAP_CVLAN_STRIPPING
= (1 << 0),
1549 /* Scatter FCS field of an incoming packet to host memory is supported.
1551 IB_RAW_PACKET_CAP_SCATTER_FCS
= (1 << 1),
1552 /* Checksum offloads are supported (for both send and receive). */
1553 IB_RAW_PACKET_CAP_IP_CSUM
= (1 << 2),
1554 /* When a packet is received for an RQ with no receive WQEs, the
1555 * packet processing is delayed.
1557 IB_RAW_PACKET_CAP_DELAY_DROP
= (1 << 3),
1571 struct ib_device
*device
;
1572 struct ib_uobject
*uobject
;
1574 void (*event_handler
)(struct ib_event
*, void *);
1578 enum ib_wq_state state
;
1579 enum ib_wq_type wq_type
;
1584 IB_WQ_FLAGS_CVLAN_STRIPPING
= 1 << 0,
1585 IB_WQ_FLAGS_SCATTER_FCS
= 1 << 1,
1586 IB_WQ_FLAGS_DELAY_DROP
= 1 << 2,
1589 struct ib_wq_init_attr
{
1591 enum ib_wq_type wq_type
;
1595 void (*event_handler
)(struct ib_event
*, void *);
1596 u32 create_flags
; /* Use enum ib_wq_flags */
1599 enum ib_wq_attr_mask
{
1600 IB_WQ_STATE
= 1 << 0,
1601 IB_WQ_CUR_STATE
= 1 << 1,
1602 IB_WQ_FLAGS
= 1 << 2,
1606 enum ib_wq_state wq_state
;
1607 enum ib_wq_state curr_wq_state
;
1608 u32 flags
; /* Use enum ib_wq_flags */
1609 u32 flags_mask
; /* Use enum ib_wq_flags */
1612 struct ib_rwq_ind_table
{
1613 struct ib_device
*device
;
1614 struct ib_uobject
*uobject
;
1617 u32 log_ind_tbl_size
;
1618 struct ib_wq
**ind_tbl
;
1621 struct ib_rwq_ind_table_init_attr
{
1622 u32 log_ind_tbl_size
;
1623 /* Each entry is a pointer to Receive Work Queue */
1624 struct ib_wq
**ind_tbl
;
1627 enum port_pkey_state
{
1628 IB_PORT_PKEY_NOT_VALID
= 0,
1629 IB_PORT_PKEY_VALID
= 1,
1630 IB_PORT_PKEY_LISTED
= 2,
1633 struct ib_qp_security
;
1635 struct ib_port_pkey
{
1636 enum port_pkey_state state
;
1639 struct list_head qp_list
;
1640 struct list_head to_error_list
;
1641 struct ib_qp_security
*sec
;
1644 struct ib_ports_pkeys
{
1645 struct ib_port_pkey main
;
1646 struct ib_port_pkey alt
;
1649 struct ib_qp_security
{
1651 struct ib_device
*dev
;
1652 /* Hold this mutex when changing port and pkey settings. */
1654 struct ib_ports_pkeys
*ports_pkeys
;
1655 /* A list of all open shared QP handles. Required to enforce security
1656 * properly for all users of a shared QP.
1658 struct list_head shared_qp_list
;
1661 atomic_t error_list_count
;
1662 struct completion error_complete
;
1663 int error_comps_pending
;
1667 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1668 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1671 struct ib_device
*device
;
1673 struct ib_cq
*send_cq
;
1674 struct ib_cq
*recv_cq
;
1677 struct list_head rdma_mrs
;
1678 struct list_head sig_mrs
;
1680 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1681 struct list_head xrcd_list
;
1683 /* count times opened, mcast attaches, flow attaches */
1685 struct list_head open_list
;
1686 struct ib_qp
*real_qp
;
1687 struct ib_uobject
*uobject
;
1688 void (*event_handler
)(struct ib_event
*, void *);
1693 enum ib_qp_type qp_type
;
1694 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1695 struct ib_qp_security
*qp_sec
;
1699 struct ib_device
*device
;
1705 unsigned int page_size
;
1708 struct ib_uobject
*uobject
; /* user */
1709 struct list_head qp_entry
; /* FR */
1714 struct ib_device
*device
;
1716 struct ib_uobject
*uobject
;
1718 enum ib_mw_type type
;
1722 struct ib_device
*device
;
1724 struct list_head list
;
1729 /* Supported steering options */
1730 enum ib_flow_attr_type
{
1731 /* steering according to rule specifications */
1732 IB_FLOW_ATTR_NORMAL
= 0x0,
1733 /* default unicast and multicast rule -
1734 * receive all Eth traffic which isn't steered to any QP
1736 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1737 /* default multicast rule -
1738 * receive all Eth multicast traffic which isn't steered to any QP
1740 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1741 /* sniffer rule - receive all port traffic */
1742 IB_FLOW_ATTR_SNIFFER
= 0x3
1745 /* Supported steering header types */
1746 enum ib_flow_spec_type
{
1748 IB_FLOW_SPEC_ETH
= 0x20,
1749 IB_FLOW_SPEC_IB
= 0x22,
1751 IB_FLOW_SPEC_IPV4
= 0x30,
1752 IB_FLOW_SPEC_IPV6
= 0x31,
1754 IB_FLOW_SPEC_TCP
= 0x40,
1755 IB_FLOW_SPEC_UDP
= 0x41,
1756 IB_FLOW_SPEC_VXLAN_TUNNEL
= 0x50,
1757 IB_FLOW_SPEC_INNER
= 0x100,
1759 IB_FLOW_SPEC_ACTION_TAG
= 0x1000,
1760 IB_FLOW_SPEC_ACTION_DROP
= 0x1001,
1762 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1763 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1765 /* Flow steering rule priority is set according to it's domain.
1766 * Lower domain value means higher priority.
1768 enum ib_flow_domain
{
1769 IB_FLOW_DOMAIN_USER
,
1770 IB_FLOW_DOMAIN_ETHTOOL
,
1773 IB_FLOW_DOMAIN_NUM
/* Must be last */
1776 enum ib_flow_flags
{
1777 IB_FLOW_ATTR_FLAGS_DONT_TRAP
= 1UL << 1, /* Continue match, no steal */
1778 IB_FLOW_ATTR_FLAGS_RESERVED
= 1UL << 2 /* Must be last */
1781 struct ib_flow_eth_filter
{
1790 struct ib_flow_spec_eth
{
1793 struct ib_flow_eth_filter val
;
1794 struct ib_flow_eth_filter mask
;
1797 struct ib_flow_ib_filter
{
1804 struct ib_flow_spec_ib
{
1807 struct ib_flow_ib_filter val
;
1808 struct ib_flow_ib_filter mask
;
1811 /* IPv4 header flags */
1812 enum ib_ipv4_flags
{
1813 IB_IPV4_DONT_FRAG
= 0x2, /* Don't enable packet fragmentation */
1814 IB_IPV4_MORE_FRAG
= 0X4 /* For All fragmented packets except the
1815 last have this flag set */
1818 struct ib_flow_ipv4_filter
{
1829 struct ib_flow_spec_ipv4
{
1832 struct ib_flow_ipv4_filter val
;
1833 struct ib_flow_ipv4_filter mask
;
1836 struct ib_flow_ipv6_filter
{
1847 struct ib_flow_spec_ipv6
{
1850 struct ib_flow_ipv6_filter val
;
1851 struct ib_flow_ipv6_filter mask
;
1854 struct ib_flow_tcp_udp_filter
{
1861 struct ib_flow_spec_tcp_udp
{
1864 struct ib_flow_tcp_udp_filter val
;
1865 struct ib_flow_tcp_udp_filter mask
;
1868 struct ib_flow_tunnel_filter
{
1873 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1874 * the tunnel_id from val has the vni value
1876 struct ib_flow_spec_tunnel
{
1879 struct ib_flow_tunnel_filter val
;
1880 struct ib_flow_tunnel_filter mask
;
1883 struct ib_flow_spec_action_tag
{
1884 enum ib_flow_spec_type type
;
1889 struct ib_flow_spec_action_drop
{
1890 enum ib_flow_spec_type type
;
1894 union ib_flow_spec
{
1899 struct ib_flow_spec_eth eth
;
1900 struct ib_flow_spec_ib ib
;
1901 struct ib_flow_spec_ipv4 ipv4
;
1902 struct ib_flow_spec_tcp_udp tcp_udp
;
1903 struct ib_flow_spec_ipv6 ipv6
;
1904 struct ib_flow_spec_tunnel tunnel
;
1905 struct ib_flow_spec_action_tag flow_tag
;
1906 struct ib_flow_spec_action_drop drop
;
1909 struct ib_flow_attr
{
1910 enum ib_flow_attr_type type
;
1916 /* Following are the optional layers according to user request
1917 * struct ib_flow_spec_xxx
1918 * struct ib_flow_spec_yyy
1924 struct ib_uobject
*uobject
;
1930 enum ib_process_mad_flags
{
1931 IB_MAD_IGNORE_MKEY
= 1,
1932 IB_MAD_IGNORE_BKEY
= 2,
1933 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1936 enum ib_mad_result
{
1937 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1938 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1939 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1940 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1943 struct ib_port_cache
{
1945 struct ib_pkey_cache
*pkey
;
1946 struct ib_gid_table
*gid
;
1948 enum ib_port_state port_state
;
1953 struct ib_event_handler event_handler
;
1954 struct ib_port_cache
*ports
;
1959 struct ib_port_immutable
{
1966 /* rdma netdev type - specifies protocol type */
1967 enum rdma_netdev_t
{
1968 RDMA_NETDEV_OPA_VNIC
,
1973 * struct rdma_netdev - rdma netdev
1974 * For cases where netstack interfacing is required.
1976 struct rdma_netdev
{
1978 struct ib_device
*hca
;
1981 /* cleanup function must be specified */
1982 void (*free_rdma_netdev
)(struct net_device
*netdev
);
1984 /* control functions */
1985 void (*set_id
)(struct net_device
*netdev
, int id
);
1987 int (*send
)(struct net_device
*dev
, struct sk_buff
*skb
,
1988 struct ib_ah
*address
, u32 dqpn
);
1990 int (*attach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
1991 union ib_gid
*gid
, u16 mlid
,
1992 int set_qkey
, u32 qkey
);
1993 int (*detach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
1994 union ib_gid
*gid
, u16 mlid
);
1997 struct ib_port_pkey_list
{
1998 /* Lock to hold while modifying the list. */
1999 spinlock_t list_lock
;
2000 struct list_head pkey_list
;
2004 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2005 struct device
*dma_device
;
2007 char name
[IB_DEVICE_NAME_MAX
];
2009 struct list_head event_handler_list
;
2010 spinlock_t event_handler_lock
;
2012 spinlock_t client_data_lock
;
2013 struct list_head core_list
;
2014 /* Access to the client_data_list is protected by the client_data_lock
2015 * spinlock and the lists_rwsem read-write semaphore */
2016 struct list_head client_data_list
;
2018 struct ib_cache cache
;
2020 * port_immutable is indexed by port number
2022 struct ib_port_immutable
*port_immutable
;
2024 int num_comp_vectors
;
2026 struct ib_port_pkey_list
*port_pkey_list
;
2028 struct iw_cm_verbs
*iwcm
;
2031 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2032 * driver initialized data. The struct is kfree()'ed by the sysfs
2033 * core when the device is removed. A lifespan of -1 in the return
2034 * struct tells the core to set a default lifespan.
2036 struct rdma_hw_stats
*(*alloc_hw_stats
)(struct ib_device
*device
,
2039 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2040 * @index - The index in the value array we wish to have updated, or
2041 * num_counters if we want all stats updated
2043 * < 0 - Error, no counters updated
2044 * index - Updated the single counter pointed to by index
2045 * num_counters - Updated all counters (will reset the timestamp
2046 * and prevent further calls for lifespan milliseconds)
2047 * Drivers are allowed to update all counters in leiu of just the
2048 * one given in index at their option
2050 int (*get_hw_stats
)(struct ib_device
*device
,
2051 struct rdma_hw_stats
*stats
,
2052 u8 port
, int index
);
2053 int (*query_device
)(struct ib_device
*device
,
2054 struct ib_device_attr
*device_attr
,
2055 struct ib_udata
*udata
);
2056 int (*query_port
)(struct ib_device
*device
,
2058 struct ib_port_attr
*port_attr
);
2059 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
2061 /* When calling get_netdev, the HW vendor's driver should return the
2062 * net device of device @device at port @port_num or NULL if such
2063 * a net device doesn't exist. The vendor driver should call dev_hold
2064 * on this net device. The HW vendor's device driver must guarantee
2065 * that this function returns NULL before the net device reaches
2066 * NETDEV_UNREGISTER_FINAL state.
2068 struct net_device
*(*get_netdev
)(struct ib_device
*device
,
2070 int (*query_gid
)(struct ib_device
*device
,
2071 u8 port_num
, int index
,
2073 /* When calling add_gid, the HW vendor's driver should
2074 * add the gid of device @device at gid index @index of
2075 * port @port_num to be @gid. Meta-info of that gid (for example,
2076 * the network device related to this gid is available
2077 * at @attr. @context allows the HW vendor driver to store extra
2078 * information together with a GID entry. The HW vendor may allocate
2079 * memory to contain this information and store it in @context when a
2080 * new GID entry is written to. Params are consistent until the next
2081 * call of add_gid or delete_gid. The function should return 0 on
2082 * success or error otherwise. The function could be called
2083 * concurrently for different ports. This function is only called
2084 * when roce_gid_table is used.
2086 int (*add_gid
)(struct ib_device
*device
,
2089 const union ib_gid
*gid
,
2090 const struct ib_gid_attr
*attr
,
2092 /* When calling del_gid, the HW vendor's driver should delete the
2093 * gid of device @device at gid index @index of port @port_num.
2094 * Upon the deletion of a GID entry, the HW vendor must free any
2095 * allocated memory. The caller will clear @context afterwards.
2096 * This function is only called when roce_gid_table is used.
2098 int (*del_gid
)(struct ib_device
*device
,
2102 int (*query_pkey
)(struct ib_device
*device
,
2103 u8 port_num
, u16 index
, u16
*pkey
);
2104 int (*modify_device
)(struct ib_device
*device
,
2105 int device_modify_mask
,
2106 struct ib_device_modify
*device_modify
);
2107 int (*modify_port
)(struct ib_device
*device
,
2108 u8 port_num
, int port_modify_mask
,
2109 struct ib_port_modify
*port_modify
);
2110 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
2111 struct ib_udata
*udata
);
2112 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
2113 int (*mmap
)(struct ib_ucontext
*context
,
2114 struct vm_area_struct
*vma
);
2115 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
2116 struct ib_ucontext
*context
,
2117 struct ib_udata
*udata
);
2118 int (*dealloc_pd
)(struct ib_pd
*pd
);
2119 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
2120 struct rdma_ah_attr
*ah_attr
,
2121 struct ib_udata
*udata
);
2122 int (*modify_ah
)(struct ib_ah
*ah
,
2123 struct rdma_ah_attr
*ah_attr
);
2124 int (*query_ah
)(struct ib_ah
*ah
,
2125 struct rdma_ah_attr
*ah_attr
);
2126 int (*destroy_ah
)(struct ib_ah
*ah
);
2127 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
2128 struct ib_srq_init_attr
*srq_init_attr
,
2129 struct ib_udata
*udata
);
2130 int (*modify_srq
)(struct ib_srq
*srq
,
2131 struct ib_srq_attr
*srq_attr
,
2132 enum ib_srq_attr_mask srq_attr_mask
,
2133 struct ib_udata
*udata
);
2134 int (*query_srq
)(struct ib_srq
*srq
,
2135 struct ib_srq_attr
*srq_attr
);
2136 int (*destroy_srq
)(struct ib_srq
*srq
);
2137 int (*post_srq_recv
)(struct ib_srq
*srq
,
2138 struct ib_recv_wr
*recv_wr
,
2139 struct ib_recv_wr
**bad_recv_wr
);
2140 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
2141 struct ib_qp_init_attr
*qp_init_attr
,
2142 struct ib_udata
*udata
);
2143 int (*modify_qp
)(struct ib_qp
*qp
,
2144 struct ib_qp_attr
*qp_attr
,
2146 struct ib_udata
*udata
);
2147 int (*query_qp
)(struct ib_qp
*qp
,
2148 struct ib_qp_attr
*qp_attr
,
2150 struct ib_qp_init_attr
*qp_init_attr
);
2151 int (*destroy_qp
)(struct ib_qp
*qp
);
2152 int (*post_send
)(struct ib_qp
*qp
,
2153 struct ib_send_wr
*send_wr
,
2154 struct ib_send_wr
**bad_send_wr
);
2155 int (*post_recv
)(struct ib_qp
*qp
,
2156 struct ib_recv_wr
*recv_wr
,
2157 struct ib_recv_wr
**bad_recv_wr
);
2158 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
2159 const struct ib_cq_init_attr
*attr
,
2160 struct ib_ucontext
*context
,
2161 struct ib_udata
*udata
);
2162 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
2164 int (*destroy_cq
)(struct ib_cq
*cq
);
2165 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
2166 struct ib_udata
*udata
);
2167 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
2169 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
2170 int (*req_notify_cq
)(struct ib_cq
*cq
,
2171 enum ib_cq_notify_flags flags
);
2172 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
2174 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
2175 int mr_access_flags
);
2176 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
2177 u64 start
, u64 length
,
2179 int mr_access_flags
,
2180 struct ib_udata
*udata
);
2181 int (*rereg_user_mr
)(struct ib_mr
*mr
,
2183 u64 start
, u64 length
,
2185 int mr_access_flags
,
2187 struct ib_udata
*udata
);
2188 int (*dereg_mr
)(struct ib_mr
*mr
);
2189 struct ib_mr
* (*alloc_mr
)(struct ib_pd
*pd
,
2190 enum ib_mr_type mr_type
,
2192 int (*map_mr_sg
)(struct ib_mr
*mr
,
2193 struct scatterlist
*sg
,
2195 unsigned int *sg_offset
);
2196 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
2197 enum ib_mw_type type
,
2198 struct ib_udata
*udata
);
2199 int (*dealloc_mw
)(struct ib_mw
*mw
);
2200 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
2201 int mr_access_flags
,
2202 struct ib_fmr_attr
*fmr_attr
);
2203 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
2204 u64
*page_list
, int list_len
,
2206 int (*unmap_fmr
)(struct list_head
*fmr_list
);
2207 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
2208 int (*attach_mcast
)(struct ib_qp
*qp
,
2211 int (*detach_mcast
)(struct ib_qp
*qp
,
2214 int (*process_mad
)(struct ib_device
*device
,
2215 int process_mad_flags
,
2217 const struct ib_wc
*in_wc
,
2218 const struct ib_grh
*in_grh
,
2219 const struct ib_mad_hdr
*in_mad
,
2221 struct ib_mad_hdr
*out_mad
,
2222 size_t *out_mad_size
,
2223 u16
*out_mad_pkey_index
);
2224 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
2225 struct ib_ucontext
*ucontext
,
2226 struct ib_udata
*udata
);
2227 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
2228 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
2232 int (*destroy_flow
)(struct ib_flow
*flow_id
);
2233 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
2234 struct ib_mr_status
*mr_status
);
2235 void (*disassociate_ucontext
)(struct ib_ucontext
*ibcontext
);
2236 void (*drain_rq
)(struct ib_qp
*qp
);
2237 void (*drain_sq
)(struct ib_qp
*qp
);
2238 int (*set_vf_link_state
)(struct ib_device
*device
, int vf
, u8 port
,
2240 int (*get_vf_config
)(struct ib_device
*device
, int vf
, u8 port
,
2241 struct ifla_vf_info
*ivf
);
2242 int (*get_vf_stats
)(struct ib_device
*device
, int vf
, u8 port
,
2243 struct ifla_vf_stats
*stats
);
2244 int (*set_vf_guid
)(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2246 struct ib_wq
* (*create_wq
)(struct ib_pd
*pd
,
2247 struct ib_wq_init_attr
*init_attr
,
2248 struct ib_udata
*udata
);
2249 int (*destroy_wq
)(struct ib_wq
*wq
);
2250 int (*modify_wq
)(struct ib_wq
*wq
,
2251 struct ib_wq_attr
*attr
,
2253 struct ib_udata
*udata
);
2254 struct ib_rwq_ind_table
* (*create_rwq_ind_table
)(struct ib_device
*device
,
2255 struct ib_rwq_ind_table_init_attr
*init_attr
,
2256 struct ib_udata
*udata
);
2257 int (*destroy_rwq_ind_table
)(struct ib_rwq_ind_table
*wq_ind_table
);
2259 * rdma netdev operation
2261 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2262 * doesn't support the specified rdma netdev type.
2264 struct net_device
*(*alloc_rdma_netdev
)(
2265 struct ib_device
*device
,
2267 enum rdma_netdev_t type
,
2269 unsigned char name_assign_type
,
2270 void (*setup
)(struct net_device
*));
2272 struct module
*owner
;
2274 struct kobject
*ports_parent
;
2275 struct list_head port_list
;
2278 IB_DEV_UNINITIALIZED
,
2284 u64 uverbs_cmd_mask
;
2285 u64 uverbs_ex_cmd_mask
;
2287 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
2293 struct ib_device_attr attrs
;
2294 struct attribute_group
*hw_stats_ag
;
2295 struct rdma_hw_stats
*hw_stats
;
2297 #ifdef CONFIG_CGROUP_RDMA
2298 struct rdmacg_device cg_device
;
2302 * The following mandatory functions are used only at device
2303 * registration. Keep functions such as these at the end of this
2304 * structure to avoid cache line misses when accessing struct ib_device
2307 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
2308 void (*get_dev_fw_str
)(struct ib_device
*, char *str
, size_t str_len
);
2313 void (*add
) (struct ib_device
*);
2314 void (*remove
)(struct ib_device
*, void *client_data
);
2316 /* Returns the net_dev belonging to this ib_client and matching the
2318 * @dev: An RDMA device that the net_dev use for communication.
2319 * @port: A physical port number on the RDMA device.
2320 * @pkey: P_Key that the net_dev uses if applicable.
2321 * @gid: A GID that the net_dev uses to communicate.
2322 * @addr: An IP address the net_dev is configured with.
2323 * @client_data: The device's client data set by ib_set_client_data().
2325 * An ib_client that implements a net_dev on top of RDMA devices
2326 * (such as IP over IB) should implement this callback, allowing the
2327 * rdma_cm module to find the right net_dev for a given request.
2329 * The caller is responsible for calling dev_put on the returned
2331 struct net_device
*(*get_net_dev_by_params
)(
2332 struct ib_device
*dev
,
2335 const union ib_gid
*gid
,
2336 const struct sockaddr
*addr
,
2338 struct list_head list
;
2341 struct ib_device
*ib_alloc_device(size_t size
);
2342 void ib_dealloc_device(struct ib_device
*device
);
2344 void ib_get_device_fw_str(struct ib_device
*device
, char *str
, size_t str_len
);
2346 int ib_register_device(struct ib_device
*device
,
2347 int (*port_callback
)(struct ib_device
*,
2348 u8
, struct kobject
*));
2349 void ib_unregister_device(struct ib_device
*device
);
2351 int ib_register_client (struct ib_client
*client
);
2352 void ib_unregister_client(struct ib_client
*client
);
2354 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
2355 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
2358 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
2360 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
2363 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
2365 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
2368 static inline bool ib_is_udata_cleared(struct ib_udata
*udata
,
2372 const void __user
*p
= udata
->inbuf
+ offset
;
2376 if (len
> USHRT_MAX
)
2379 buf
= memdup_user(p
, len
);
2383 ret
= !memchr_inv(buf
, 0, len
);
2389 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2390 * contains all required attributes and no attributes not allowed for
2391 * the given QP state transition.
2392 * @cur_state: Current QP state
2393 * @next_state: Next QP state
2395 * @mask: Mask of supplied QP attributes
2396 * @ll : link layer of port
2398 * This function is a helper function that a low-level driver's
2399 * modify_qp method can use to validate the consumer's input. It
2400 * checks that cur_state and next_state are valid QP states, that a
2401 * transition from cur_state to next_state is allowed by the IB spec,
2402 * and that the attribute mask supplied is allowed for the transition.
2404 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
2405 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
2406 enum rdma_link_layer ll
);
2408 int ib_register_event_handler (struct ib_event_handler
*event_handler
);
2409 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
2410 void ib_dispatch_event(struct ib_event
*event
);
2412 int ib_query_port(struct ib_device
*device
,
2413 u8 port_num
, struct ib_port_attr
*port_attr
);
2415 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
2419 * rdma_cap_ib_switch - Check if the device is IB switch
2420 * @device: Device to check
2422 * Device driver is responsible for setting is_switch bit on
2423 * in ib_device structure at init time.
2425 * Return: true if the device is IB switch.
2427 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
2429 return device
->is_switch
;
2433 * rdma_start_port - Return the first valid port number for the device
2436 * @device: Device to be checked
2438 * Return start port number
2440 static inline u8
rdma_start_port(const struct ib_device
*device
)
2442 return rdma_cap_ib_switch(device
) ? 0 : 1;
2446 * rdma_end_port - Return the last valid port number for the device
2449 * @device: Device to be checked
2451 * Return last port number
2453 static inline u8
rdma_end_port(const struct ib_device
*device
)
2455 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
2458 static inline int rdma_is_port_valid(const struct ib_device
*device
,
2461 return (port
>= rdma_start_port(device
) &&
2462 port
<= rdma_end_port(device
));
2465 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
2467 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
2470 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
2472 return device
->port_immutable
[port_num
].core_cap_flags
&
2473 (RDMA_CORE_CAP_PROT_ROCE
| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
);
2476 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device
*device
, u8 port_num
)
2478 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
;
2481 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device
*device
, u8 port_num
)
2483 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
2486 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
2488 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
2491 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
2493 return rdma_protocol_ib(device
, port_num
) ||
2494 rdma_protocol_roce(device
, port_num
);
2497 static inline bool rdma_protocol_raw_packet(const struct ib_device
*device
, u8 port_num
)
2499 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_RAW_PACKET
;
2502 static inline bool rdma_protocol_usnic(const struct ib_device
*device
, u8 port_num
)
2504 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_USNIC
;
2508 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2509 * Management Datagrams.
2510 * @device: Device to check
2511 * @port_num: Port number to check
2513 * Management Datagrams (MAD) are a required part of the InfiniBand
2514 * specification and are supported on all InfiniBand devices. A slightly
2515 * extended version are also supported on OPA interfaces.
2517 * Return: true if the port supports sending/receiving of MAD packets.
2519 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
2521 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
2525 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2526 * Management Datagrams.
2527 * @device: Device to check
2528 * @port_num: Port number to check
2530 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2531 * datagrams with their own versions. These OPA MADs share many but not all of
2532 * the characteristics of InfiniBand MADs.
2534 * OPA MADs differ in the following ways:
2536 * 1) MADs are variable size up to 2K
2537 * IBTA defined MADs remain fixed at 256 bytes
2538 * 2) OPA SMPs must carry valid PKeys
2539 * 3) OPA SMP packets are a different format
2541 * Return: true if the port supports OPA MAD packet formats.
2543 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
2545 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
2546 == RDMA_CORE_CAP_OPA_MAD
;
2550 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2551 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2552 * @device: Device to check
2553 * @port_num: Port number to check
2555 * Each InfiniBand node is required to provide a Subnet Management Agent
2556 * that the subnet manager can access. Prior to the fabric being fully
2557 * configured by the subnet manager, the SMA is accessed via a well known
2558 * interface called the Subnet Management Interface (SMI). This interface
2559 * uses directed route packets to communicate with the SM to get around the
2560 * chicken and egg problem of the SM needing to know what's on the fabric
2561 * in order to configure the fabric, and needing to configure the fabric in
2562 * order to send packets to the devices on the fabric. These directed
2563 * route packets do not need the fabric fully configured in order to reach
2564 * their destination. The SMI is the only method allowed to send
2565 * directed route packets on an InfiniBand fabric.
2567 * Return: true if the port provides an SMI.
2569 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
2571 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
2575 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2576 * Communication Manager.
2577 * @device: Device to check
2578 * @port_num: Port number to check
2580 * The InfiniBand Communication Manager is one of many pre-defined General
2581 * Service Agents (GSA) that are accessed via the General Service
2582 * Interface (GSI). It's role is to facilitate establishment of connections
2583 * between nodes as well as other management related tasks for established
2586 * Return: true if the port supports an IB CM (this does not guarantee that
2587 * a CM is actually running however).
2589 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
2591 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2595 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2596 * Communication Manager.
2597 * @device: Device to check
2598 * @port_num: Port number to check
2600 * Similar to above, but specific to iWARP connections which have a different
2601 * managment protocol than InfiniBand.
2603 * Return: true if the port supports an iWARP CM (this does not guarantee that
2604 * a CM is actually running however).
2606 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2608 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2612 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2613 * Subnet Administration.
2614 * @device: Device to check
2615 * @port_num: Port number to check
2617 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2618 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2619 * fabrics, devices should resolve routes to other hosts by contacting the
2620 * SA to query the proper route.
2622 * Return: true if the port should act as a client to the fabric Subnet
2623 * Administration interface. This does not imply that the SA service is
2626 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2628 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2632 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2634 * @device: Device to check
2635 * @port_num: Port number to check
2637 * InfiniBand multicast registration is more complex than normal IPv4 or
2638 * IPv6 multicast registration. Each Host Channel Adapter must register
2639 * with the Subnet Manager when it wishes to join a multicast group. It
2640 * should do so only once regardless of how many queue pairs it subscribes
2641 * to this group. And it should leave the group only after all queue pairs
2642 * attached to the group have been detached.
2644 * Return: true if the port must undertake the additional adminstrative
2645 * overhead of registering/unregistering with the SM and tracking of the
2646 * total number of queue pairs attached to the multicast group.
2648 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2650 return rdma_cap_ib_sa(device
, port_num
);
2654 * rdma_cap_af_ib - Check if the port of device has the capability
2655 * Native Infiniband Address.
2656 * @device: Device to check
2657 * @port_num: Port number to check
2659 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2660 * GID. RoCE uses a different mechanism, but still generates a GID via
2661 * a prescribed mechanism and port specific data.
2663 * Return: true if the port uses a GID address to identify devices on the
2666 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2668 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2672 * rdma_cap_eth_ah - Check if the port of device has the capability
2673 * Ethernet Address Handle.
2674 * @device: Device to check
2675 * @port_num: Port number to check
2677 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2678 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2679 * port. Normally, packet headers are generated by the sending host
2680 * adapter, but when sending connectionless datagrams, we must manually
2681 * inject the proper headers for the fabric we are communicating over.
2683 * Return: true if we are running as a RoCE port and must force the
2684 * addition of a Global Route Header built from our Ethernet Address
2685 * Handle into our header list for connectionless packets.
2687 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2689 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2693 * rdma_cap_opa_ah - Check if the port of device supports
2694 * OPA Address handles
2695 * @device: Device to check
2696 * @port_num: Port number to check
2698 * Return: true if we are running on an OPA device which supports
2699 * the extended OPA addressing.
2701 static inline bool rdma_cap_opa_ah(struct ib_device
*device
, u8 port_num
)
2703 return (device
->port_immutable
[port_num
].core_cap_flags
&
2704 RDMA_CORE_CAP_OPA_AH
) == RDMA_CORE_CAP_OPA_AH
;
2708 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2711 * @port_num: Port number
2713 * This MAD size includes the MAD headers and MAD payload. No other headers
2716 * Return the max MAD size required by the Port. Will return 0 if the port
2717 * does not support MADs
2719 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2721 return device
->port_immutable
[port_num
].max_mad_size
;
2725 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2726 * @device: Device to check
2727 * @port_num: Port number to check
2729 * RoCE GID table mechanism manages the various GIDs for a device.
2731 * NOTE: if allocating the port's GID table has failed, this call will still
2732 * return true, but any RoCE GID table API will fail.
2734 * Return: true if the port uses RoCE GID table mechanism in order to manage
2737 static inline bool rdma_cap_roce_gid_table(const struct ib_device
*device
,
2740 return rdma_protocol_roce(device
, port_num
) &&
2741 device
->add_gid
&& device
->del_gid
;
2745 * Check if the device supports READ W/ INVALIDATE.
2747 static inline bool rdma_cap_read_inv(struct ib_device
*dev
, u32 port_num
)
2750 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2751 * has support for it yet.
2753 return rdma_protocol_iwarp(dev
, port_num
);
2756 int ib_query_gid(struct ib_device
*device
,
2757 u8 port_num
, int index
, union ib_gid
*gid
,
2758 struct ib_gid_attr
*attr
);
2760 int ib_set_vf_link_state(struct ib_device
*device
, int vf
, u8 port
,
2762 int ib_get_vf_config(struct ib_device
*device
, int vf
, u8 port
,
2763 struct ifla_vf_info
*info
);
2764 int ib_get_vf_stats(struct ib_device
*device
, int vf
, u8 port
,
2765 struct ifla_vf_stats
*stats
);
2766 int ib_set_vf_guid(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2769 int ib_query_pkey(struct ib_device
*device
,
2770 u8 port_num
, u16 index
, u16
*pkey
);
2772 int ib_modify_device(struct ib_device
*device
,
2773 int device_modify_mask
,
2774 struct ib_device_modify
*device_modify
);
2776 int ib_modify_port(struct ib_device
*device
,
2777 u8 port_num
, int port_modify_mask
,
2778 struct ib_port_modify
*port_modify
);
2780 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2781 enum ib_gid_type gid_type
, struct net_device
*ndev
,
2782 u8
*port_num
, u16
*index
);
2784 int ib_find_pkey(struct ib_device
*device
,
2785 u8 port_num
, u16 pkey
, u16
*index
);
2789 * Create a memory registration for all memory in the system and place
2790 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2791 * ULPs to avoid the overhead of dynamic MRs.
2793 * This flag is generally considered unsafe and must only be used in
2794 * extremly trusted environments. Every use of it will log a warning
2795 * in the kernel log.
2797 IB_PD_UNSAFE_GLOBAL_RKEY
= 0x01,
2800 struct ib_pd
*__ib_alloc_pd(struct ib_device
*device
, unsigned int flags
,
2801 const char *caller
);
2802 #define ib_alloc_pd(device, flags) \
2803 __ib_alloc_pd((device), (flags), __func__)
2804 void ib_dealloc_pd(struct ib_pd
*pd
);
2807 * rdma_create_ah - Creates an address handle for the given address vector.
2808 * @pd: The protection domain associated with the address handle.
2809 * @ah_attr: The attributes of the address vector.
2811 * The address handle is used to reference a local or global destination
2812 * in all UD QP post sends.
2814 struct ib_ah
*rdma_create_ah(struct ib_pd
*pd
, struct rdma_ah_attr
*ah_attr
);
2817 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2819 * @hdr: the L3 header to parse
2820 * @net_type: type of header to parse
2821 * @sgid: place to store source gid
2822 * @dgid: place to store destination gid
2824 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr
*hdr
,
2825 enum rdma_network_type net_type
,
2826 union ib_gid
*sgid
, union ib_gid
*dgid
);
2829 * ib_get_rdma_header_version - Get the header version
2830 * @hdr: the L3 header to parse
2832 int ib_get_rdma_header_version(const union rdma_network_hdr
*hdr
);
2835 * ib_init_ah_from_wc - Initializes address handle attributes from a
2837 * @device: Device on which the received message arrived.
2838 * @port_num: Port on which the received message arrived.
2839 * @wc: Work completion associated with the received message.
2840 * @grh: References the received global route header. This parameter is
2841 * ignored unless the work completion indicates that the GRH is valid.
2842 * @ah_attr: Returned attributes that can be used when creating an address
2843 * handle for replying to the message.
2845 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2846 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2847 struct rdma_ah_attr
*ah_attr
);
2850 * ib_create_ah_from_wc - Creates an address handle associated with the
2851 * sender of the specified work completion.
2852 * @pd: The protection domain associated with the address handle.
2853 * @wc: Work completion information associated with a received message.
2854 * @grh: References the received global route header. This parameter is
2855 * ignored unless the work completion indicates that the GRH is valid.
2856 * @port_num: The outbound port number to associate with the address.
2858 * The address handle is used to reference a local or global destination
2859 * in all UD QP post sends.
2861 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2862 const struct ib_grh
*grh
, u8 port_num
);
2865 * rdma_modify_ah - Modifies the address vector associated with an address
2867 * @ah: The address handle to modify.
2868 * @ah_attr: The new address vector attributes to associate with the
2871 int rdma_modify_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2874 * rdma_query_ah - Queries the address vector associated with an address
2876 * @ah: The address handle to query.
2877 * @ah_attr: The address vector attributes associated with the address
2880 int rdma_query_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2883 * rdma_destroy_ah - Destroys an address handle.
2884 * @ah: The address handle to destroy.
2886 int rdma_destroy_ah(struct ib_ah
*ah
);
2889 * ib_create_srq - Creates a SRQ associated with the specified protection
2891 * @pd: The protection domain associated with the SRQ.
2892 * @srq_init_attr: A list of initial attributes required to create the
2893 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2894 * the actual capabilities of the created SRQ.
2896 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2897 * requested size of the SRQ, and set to the actual values allocated
2898 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2899 * will always be at least as large as the requested values.
2901 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2902 struct ib_srq_init_attr
*srq_init_attr
);
2905 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2906 * @srq: The SRQ to modify.
2907 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2908 * the current values of selected SRQ attributes are returned.
2909 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2910 * are being modified.
2912 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2913 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2914 * the number of receives queued drops below the limit.
2916 int ib_modify_srq(struct ib_srq
*srq
,
2917 struct ib_srq_attr
*srq_attr
,
2918 enum ib_srq_attr_mask srq_attr_mask
);
2921 * ib_query_srq - Returns the attribute list and current values for the
2923 * @srq: The SRQ to query.
2924 * @srq_attr: The attributes of the specified SRQ.
2926 int ib_query_srq(struct ib_srq
*srq
,
2927 struct ib_srq_attr
*srq_attr
);
2930 * ib_destroy_srq - Destroys the specified SRQ.
2931 * @srq: The SRQ to destroy.
2933 int ib_destroy_srq(struct ib_srq
*srq
);
2936 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2937 * @srq: The SRQ to post the work request on.
2938 * @recv_wr: A list of work requests to post on the receive queue.
2939 * @bad_recv_wr: On an immediate failure, this parameter will reference
2940 * the work request that failed to be posted on the QP.
2942 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
2943 struct ib_recv_wr
*recv_wr
,
2944 struct ib_recv_wr
**bad_recv_wr
)
2946 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
2950 * ib_create_qp - Creates a QP associated with the specified protection
2952 * @pd: The protection domain associated with the QP.
2953 * @qp_init_attr: A list of initial attributes required to create the
2954 * QP. If QP creation succeeds, then the attributes are updated to
2955 * the actual capabilities of the created QP.
2957 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
2958 struct ib_qp_init_attr
*qp_init_attr
);
2961 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
2962 * @qp: The QP to modify.
2963 * @attr: On input, specifies the QP attributes to modify. On output,
2964 * the current values of selected QP attributes are returned.
2965 * @attr_mask: A bit-mask used to specify which attributes of the QP
2966 * are being modified.
2967 * @udata: pointer to user's input output buffer information
2968 * are being modified.
2969 * It returns 0 on success and returns appropriate error code on error.
2971 int ib_modify_qp_with_udata(struct ib_qp
*qp
,
2972 struct ib_qp_attr
*attr
,
2974 struct ib_udata
*udata
);
2977 * ib_modify_qp - Modifies the attributes for the specified QP and then
2978 * transitions the QP to the given state.
2979 * @qp: The QP to modify.
2980 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2981 * the current values of selected QP attributes are returned.
2982 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2983 * are being modified.
2985 int ib_modify_qp(struct ib_qp
*qp
,
2986 struct ib_qp_attr
*qp_attr
,
2990 * ib_query_qp - Returns the attribute list and current values for the
2992 * @qp: The QP to query.
2993 * @qp_attr: The attributes of the specified QP.
2994 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2995 * @qp_init_attr: Additional attributes of the selected QP.
2997 * The qp_attr_mask may be used to limit the query to gathering only the
2998 * selected attributes.
3000 int ib_query_qp(struct ib_qp
*qp
,
3001 struct ib_qp_attr
*qp_attr
,
3003 struct ib_qp_init_attr
*qp_init_attr
);
3006 * ib_destroy_qp - Destroys the specified QP.
3007 * @qp: The QP to destroy.
3009 int ib_destroy_qp(struct ib_qp
*qp
);
3012 * ib_open_qp - Obtain a reference to an existing sharable QP.
3013 * @xrcd - XRC domain
3014 * @qp_open_attr: Attributes identifying the QP to open.
3016 * Returns a reference to a sharable QP.
3018 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
3019 struct ib_qp_open_attr
*qp_open_attr
);
3022 * ib_close_qp - Release an external reference to a QP.
3023 * @qp: The QP handle to release
3025 * The opened QP handle is released by the caller. The underlying
3026 * shared QP is not destroyed until all internal references are released.
3028 int ib_close_qp(struct ib_qp
*qp
);
3031 * ib_post_send - Posts a list of work requests to the send queue of
3033 * @qp: The QP to post the work request on.
3034 * @send_wr: A list of work requests to post on the send queue.
3035 * @bad_send_wr: On an immediate failure, this parameter will reference
3036 * the work request that failed to be posted on the QP.
3038 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3039 * error is returned, the QP state shall not be affected,
3040 * ib_post_send() will return an immediate error after queueing any
3041 * earlier work requests in the list.
3043 static inline int ib_post_send(struct ib_qp
*qp
,
3044 struct ib_send_wr
*send_wr
,
3045 struct ib_send_wr
**bad_send_wr
)
3047 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
3051 * ib_post_recv - Posts a list of work requests to the receive queue of
3053 * @qp: The QP to post the work request on.
3054 * @recv_wr: A list of work requests to post on the receive queue.
3055 * @bad_recv_wr: On an immediate failure, this parameter will reference
3056 * the work request that failed to be posted on the QP.
3058 static inline int ib_post_recv(struct ib_qp
*qp
,
3059 struct ib_recv_wr
*recv_wr
,
3060 struct ib_recv_wr
**bad_recv_wr
)
3062 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
3065 struct ib_cq
*ib_alloc_cq(struct ib_device
*dev
, void *private,
3066 int nr_cqe
, int comp_vector
, enum ib_poll_context poll_ctx
);
3067 void ib_free_cq(struct ib_cq
*cq
);
3068 int ib_process_cq_direct(struct ib_cq
*cq
, int budget
);
3071 * ib_create_cq - Creates a CQ on the specified device.
3072 * @device: The device on which to create the CQ.
3073 * @comp_handler: A user-specified callback that is invoked when a
3074 * completion event occurs on the CQ.
3075 * @event_handler: A user-specified callback that is invoked when an
3076 * asynchronous event not associated with a completion occurs on the CQ.
3077 * @cq_context: Context associated with the CQ returned to the user via
3078 * the associated completion and event handlers.
3079 * @cq_attr: The attributes the CQ should be created upon.
3081 * Users can examine the cq structure to determine the actual CQ size.
3083 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
3084 ib_comp_handler comp_handler
,
3085 void (*event_handler
)(struct ib_event
*, void *),
3087 const struct ib_cq_init_attr
*cq_attr
);
3090 * ib_resize_cq - Modifies the capacity of the CQ.
3091 * @cq: The CQ to resize.
3092 * @cqe: The minimum size of the CQ.
3094 * Users can examine the cq structure to determine the actual CQ size.
3096 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
3099 * ib_modify_cq - Modifies moderation params of the CQ
3100 * @cq: The CQ to modify.
3101 * @cq_count: number of CQEs that will trigger an event
3102 * @cq_period: max period of time in usec before triggering an event
3105 int ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
3108 * ib_destroy_cq - Destroys the specified CQ.
3109 * @cq: The CQ to destroy.
3111 int ib_destroy_cq(struct ib_cq
*cq
);
3114 * ib_poll_cq - poll a CQ for completion(s)
3115 * @cq:the CQ being polled
3116 * @num_entries:maximum number of completions to return
3117 * @wc:array of at least @num_entries &struct ib_wc where completions
3120 * Poll a CQ for (possibly multiple) completions. If the return value
3121 * is < 0, an error occurred. If the return value is >= 0, it is the
3122 * number of completions returned. If the return value is
3123 * non-negative and < num_entries, then the CQ was emptied.
3125 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
3128 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
3132 * ib_peek_cq - Returns the number of unreaped completions currently
3133 * on the specified CQ.
3134 * @cq: The CQ to peek.
3135 * @wc_cnt: A minimum number of unreaped completions to check for.
3137 * If the number of unreaped completions is greater than or equal to wc_cnt,
3138 * this function returns wc_cnt, otherwise, it returns the actual number of
3139 * unreaped completions.
3141 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
3144 * ib_req_notify_cq - Request completion notification on a CQ.
3145 * @cq: The CQ to generate an event for.
3147 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3148 * to request an event on the next solicited event or next work
3149 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3150 * may also be |ed in to request a hint about missed events, as
3154 * < 0 means an error occurred while requesting notification
3155 * == 0 means notification was requested successfully, and if
3156 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3157 * were missed and it is safe to wait for another event. In
3158 * this case is it guaranteed that any work completions added
3159 * to the CQ since the last CQ poll will trigger a completion
3160 * notification event.
3161 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3162 * in. It means that the consumer must poll the CQ again to
3163 * make sure it is empty to avoid missing an event because of a
3164 * race between requesting notification and an entry being
3165 * added to the CQ. This return value means it is possible
3166 * (but not guaranteed) that a work completion has been added
3167 * to the CQ since the last poll without triggering a
3168 * completion notification event.
3170 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
3171 enum ib_cq_notify_flags flags
)
3173 return cq
->device
->req_notify_cq(cq
, flags
);
3177 * ib_req_ncomp_notif - Request completion notification when there are
3178 * at least the specified number of unreaped completions on the CQ.
3179 * @cq: The CQ to generate an event for.
3180 * @wc_cnt: The number of unreaped completions that should be on the
3181 * CQ before an event is generated.
3183 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
3185 return cq
->device
->req_ncomp_notif
?
3186 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
3191 * ib_dma_mapping_error - check a DMA addr for error
3192 * @dev: The device for which the dma_addr was created
3193 * @dma_addr: The DMA address to check
3195 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
3197 return dma_mapping_error(dev
->dma_device
, dma_addr
);
3201 * ib_dma_map_single - Map a kernel virtual address to DMA address
3202 * @dev: The device for which the dma_addr is to be created
3203 * @cpu_addr: The kernel virtual address
3204 * @size: The size of the region in bytes
3205 * @direction: The direction of the DMA
3207 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
3208 void *cpu_addr
, size_t size
,
3209 enum dma_data_direction direction
)
3211 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
3215 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3216 * @dev: The device for which the DMA address was created
3217 * @addr: The DMA address
3218 * @size: The size of the region in bytes
3219 * @direction: The direction of the DMA
3221 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
3222 u64 addr
, size_t size
,
3223 enum dma_data_direction direction
)
3225 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
3229 * ib_dma_map_page - Map a physical page to DMA address
3230 * @dev: The device for which the dma_addr is to be created
3231 * @page: The page to be mapped
3232 * @offset: The offset within the page
3233 * @size: The size of the region in bytes
3234 * @direction: The direction of the DMA
3236 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
3238 unsigned long offset
,
3240 enum dma_data_direction direction
)
3242 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
3246 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3247 * @dev: The device for which the DMA address was created
3248 * @addr: The DMA address
3249 * @size: The size of the region in bytes
3250 * @direction: The direction of the DMA
3252 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
3253 u64 addr
, size_t size
,
3254 enum dma_data_direction direction
)
3256 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
3260 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3261 * @dev: The device for which the DMA addresses are to be created
3262 * @sg: The array of scatter/gather entries
3263 * @nents: The number of scatter/gather entries
3264 * @direction: The direction of the DMA
3266 static inline int ib_dma_map_sg(struct ib_device
*dev
,
3267 struct scatterlist
*sg
, int nents
,
3268 enum dma_data_direction direction
)
3270 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
3274 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3275 * @dev: The device for which the DMA addresses were created
3276 * @sg: The array of scatter/gather entries
3277 * @nents: The number of scatter/gather entries
3278 * @direction: The direction of the DMA
3280 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
3281 struct scatterlist
*sg
, int nents
,
3282 enum dma_data_direction direction
)
3284 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
3287 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
3288 struct scatterlist
*sg
, int nents
,
3289 enum dma_data_direction direction
,
3290 unsigned long dma_attrs
)
3292 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
,
3296 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
3297 struct scatterlist
*sg
, int nents
,
3298 enum dma_data_direction direction
,
3299 unsigned long dma_attrs
)
3301 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, dma_attrs
);
3304 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3305 * @dev: The device for which the DMA addresses were created
3306 * @sg: The scatter/gather entry
3308 * Note: this function is obsolete. To do: change all occurrences of
3309 * ib_sg_dma_address() into sg_dma_address().
3311 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
3312 struct scatterlist
*sg
)
3314 return sg_dma_address(sg
);
3318 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3319 * @dev: The device for which the DMA addresses were created
3320 * @sg: The scatter/gather entry
3322 * Note: this function is obsolete. To do: change all occurrences of
3323 * ib_sg_dma_len() into sg_dma_len().
3325 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
3326 struct scatterlist
*sg
)
3328 return sg_dma_len(sg
);
3332 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3333 * @dev: The device for which the DMA address was created
3334 * @addr: The DMA address
3335 * @size: The size of the region in bytes
3336 * @dir: The direction of the DMA
3338 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
3341 enum dma_data_direction dir
)
3343 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
3347 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3348 * @dev: The device for which the DMA address was created
3349 * @addr: The DMA address
3350 * @size: The size of the region in bytes
3351 * @dir: The direction of the DMA
3353 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
3356 enum dma_data_direction dir
)
3358 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
3362 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3363 * @dev: The device for which the DMA address is requested
3364 * @size: The size of the region to allocate in bytes
3365 * @dma_handle: A pointer for returning the DMA address of the region
3366 * @flag: memory allocator flags
3368 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
3370 dma_addr_t
*dma_handle
,
3373 return dma_alloc_coherent(dev
->dma_device
, size
, dma_handle
, flag
);
3377 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3378 * @dev: The device for which the DMA addresses were allocated
3379 * @size: The size of the region
3380 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3381 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3383 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
3384 size_t size
, void *cpu_addr
,
3385 dma_addr_t dma_handle
)
3387 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
3391 * ib_dereg_mr - Deregisters a memory region and removes it from the
3392 * HCA translation table.
3393 * @mr: The memory region to deregister.
3395 * This function can fail, if the memory region has memory windows bound to it.
3397 int ib_dereg_mr(struct ib_mr
*mr
);
3399 struct ib_mr
*ib_alloc_mr(struct ib_pd
*pd
,
3400 enum ib_mr_type mr_type
,
3404 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3406 * @mr - struct ib_mr pointer to be updated.
3407 * @newkey - new key to be used.
3409 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
3411 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
3412 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
3416 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3417 * for calculating a new rkey for type 2 memory windows.
3418 * @rkey - the rkey to increment.
3420 static inline u32
ib_inc_rkey(u32 rkey
)
3422 const u32 mask
= 0x000000ff;
3423 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
3427 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3428 * @pd: The protection domain associated with the unmapped region.
3429 * @mr_access_flags: Specifies the memory access rights.
3430 * @fmr_attr: Attributes of the unmapped region.
3432 * A fast memory region must be mapped before it can be used as part of
3435 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
3436 int mr_access_flags
,
3437 struct ib_fmr_attr
*fmr_attr
);
3440 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3441 * @fmr: The fast memory region to associate with the pages.
3442 * @page_list: An array of physical pages to map to the fast memory region.
3443 * @list_len: The number of pages in page_list.
3444 * @iova: The I/O virtual address to use with the mapped region.
3446 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
3447 u64
*page_list
, int list_len
,
3450 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
3454 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3455 * @fmr_list: A linked list of fast memory regions to unmap.
3457 int ib_unmap_fmr(struct list_head
*fmr_list
);
3460 * ib_dealloc_fmr - Deallocates a fast memory region.
3461 * @fmr: The fast memory region to deallocate.
3463 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
3466 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3467 * @qp: QP to attach to the multicast group. The QP must be type
3469 * @gid: Multicast group GID.
3470 * @lid: Multicast group LID in host byte order.
3472 * In order to send and receive multicast packets, subnet
3473 * administration must have created the multicast group and configured
3474 * the fabric appropriately. The port associated with the specified
3475 * QP must also be a member of the multicast group.
3477 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3480 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3481 * @qp: QP to detach from the multicast group.
3482 * @gid: Multicast group GID.
3483 * @lid: Multicast group LID in host byte order.
3485 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3488 * ib_alloc_xrcd - Allocates an XRC domain.
3489 * @device: The device on which to allocate the XRC domain.
3491 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3494 * ib_dealloc_xrcd - Deallocates an XRC domain.
3495 * @xrcd: The XRC domain to deallocate.
3497 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3499 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3500 struct ib_flow_attr
*flow_attr
, int domain
);
3501 int ib_destroy_flow(struct ib_flow
*flow_id
);
3503 static inline int ib_check_mr_access(int flags
)
3506 * Local write permission is required if remote write or
3507 * remote atomic permission is also requested.
3509 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3510 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3517 * ib_check_mr_status: lightweight check of MR status.
3518 * This routine may provide status checks on a selected
3519 * ib_mr. first use is for signature status check.
3521 * @mr: A memory region.
3522 * @check_mask: Bitmask of which checks to perform from
3523 * ib_mr_status_check enumeration.
3524 * @mr_status: The container of relevant status checks.
3525 * failed checks will be indicated in the status bitmask
3526 * and the relevant info shall be in the error item.
3528 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3529 struct ib_mr_status
*mr_status
);
3531 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3532 u16 pkey
, const union ib_gid
*gid
,
3533 const struct sockaddr
*addr
);
3534 struct ib_wq
*ib_create_wq(struct ib_pd
*pd
,
3535 struct ib_wq_init_attr
*init_attr
);
3536 int ib_destroy_wq(struct ib_wq
*wq
);
3537 int ib_modify_wq(struct ib_wq
*wq
, struct ib_wq_attr
*attr
,
3539 struct ib_rwq_ind_table
*ib_create_rwq_ind_table(struct ib_device
*device
,
3540 struct ib_rwq_ind_table_init_attr
*
3541 wq_ind_table_init_attr
);
3542 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table
*wq_ind_table
);
3544 int ib_map_mr_sg(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3545 unsigned int *sg_offset
, unsigned int page_size
);
3548 ib_map_mr_sg_zbva(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3549 unsigned int *sg_offset
, unsigned int page_size
)
3553 n
= ib_map_mr_sg(mr
, sg
, sg_nents
, sg_offset
, page_size
);
3559 int ib_sg_to_pages(struct ib_mr
*mr
, struct scatterlist
*sgl
, int sg_nents
,
3560 unsigned int *sg_offset
, int (*set_page
)(struct ib_mr
*, u64
));
3562 void ib_drain_rq(struct ib_qp
*qp
);
3563 void ib_drain_sq(struct ib_qp
*qp
);
3564 void ib_drain_qp(struct ib_qp
*qp
);
3566 int ib_resolve_eth_dmac(struct ib_device
*device
,
3567 struct rdma_ah_attr
*ah_attr
);
3568 int ib_get_eth_speed(struct ib_device
*dev
, u8 port_num
, u8
*speed
, u8
*width
);
3570 static inline u8
*rdma_ah_retrieve_dmac(struct rdma_ah_attr
*attr
)
3572 if (attr
->type
== RDMA_AH_ATTR_TYPE_ROCE
)
3573 return attr
->roce
.dmac
;
3577 static inline void rdma_ah_set_dlid(struct rdma_ah_attr
*attr
, u32 dlid
)
3579 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3580 attr
->ib
.dlid
= (u16
)dlid
;
3581 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3582 attr
->opa
.dlid
= dlid
;
3585 static inline u32
rdma_ah_get_dlid(const struct rdma_ah_attr
*attr
)
3587 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3588 return attr
->ib
.dlid
;
3589 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3590 return attr
->opa
.dlid
;
3594 static inline void rdma_ah_set_sl(struct rdma_ah_attr
*attr
, u8 sl
)
3599 static inline u8
rdma_ah_get_sl(const struct rdma_ah_attr
*attr
)
3604 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr
*attr
,
3607 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3608 attr
->ib
.src_path_bits
= src_path_bits
;
3609 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3610 attr
->opa
.src_path_bits
= src_path_bits
;
3613 static inline u8
rdma_ah_get_path_bits(const struct rdma_ah_attr
*attr
)
3615 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3616 return attr
->ib
.src_path_bits
;
3617 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3618 return attr
->opa
.src_path_bits
;
3622 static inline void rdma_ah_set_port_num(struct rdma_ah_attr
*attr
, u8 port_num
)
3624 attr
->port_num
= port_num
;
3627 static inline u8
rdma_ah_get_port_num(const struct rdma_ah_attr
*attr
)
3629 return attr
->port_num
;
3632 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr
*attr
,
3635 attr
->static_rate
= static_rate
;
3638 static inline u8
rdma_ah_get_static_rate(const struct rdma_ah_attr
*attr
)
3640 return attr
->static_rate
;
3643 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr
*attr
,
3644 enum ib_ah_flags flag
)
3646 attr
->ah_flags
= flag
;
3649 static inline enum ib_ah_flags
3650 rdma_ah_get_ah_flags(const struct rdma_ah_attr
*attr
)
3652 return attr
->ah_flags
;
3655 static inline const struct ib_global_route
3656 *rdma_ah_read_grh(const struct rdma_ah_attr
*attr
)
3661 /*To retrieve and modify the grh */
3662 static inline struct ib_global_route
3663 *rdma_ah_retrieve_grh(struct rdma_ah_attr
*attr
)
3668 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr
*attr
, void *dgid
)
3670 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3672 memcpy(grh
->dgid
.raw
, dgid
, sizeof(grh
->dgid
));
3675 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr
*attr
,
3678 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3680 grh
->dgid
.global
.subnet_prefix
= prefix
;
3683 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr
*attr
,
3686 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3688 grh
->dgid
.global
.interface_id
= if_id
;
3691 static inline void rdma_ah_set_grh(struct rdma_ah_attr
*attr
,
3692 union ib_gid
*dgid
, u32 flow_label
,
3693 u8 sgid_index
, u8 hop_limit
,
3696 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3698 attr
->ah_flags
= IB_AH_GRH
;
3701 grh
->flow_label
= flow_label
;
3702 grh
->sgid_index
= sgid_index
;
3703 grh
->hop_limit
= hop_limit
;
3704 grh
->traffic_class
= traffic_class
;
3708 static inline enum rdma_ah_attr_type
rdma_ah_find_type(struct ib_device
*dev
,
3711 if ((rdma_protocol_roce(dev
, port_num
)) ||
3712 (rdma_protocol_iwarp(dev
, port_num
)))
3713 return RDMA_AH_ATTR_TYPE_ROCE
;
3714 else if ((rdma_protocol_ib(dev
, port_num
)) &&
3715 (rdma_cap_opa_ah(dev
, port_num
)))
3716 return RDMA_AH_ATTR_TYPE_OPA
;
3718 return RDMA_AH_ATTR_TYPE_IB
;
3720 #endif /* IB_VERBS_H */