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)
583 struct ib_port_modify {
584 u32 set_port_cap_mask;
585 u32 clr_port_cap_mask;
593 IB_EVENT_QP_ACCESS_ERR,
597 IB_EVENT_PATH_MIG_ERR,
598 IB_EVENT_DEVICE_FATAL,
599 IB_EVENT_PORT_ACTIVE,
602 IB_EVENT_PKEY_CHANGE,
605 IB_EVENT_SRQ_LIMIT_REACHED,
606 IB_EVENT_QP_LAST_WQE_REACHED,
607 IB_EVENT_CLIENT_REREGISTER,
612 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
615 struct ib_device *device;
623 enum ib_event_type event;
626 struct ib_event_handler {
627 struct ib_device *device;
628 void (*handler)(struct ib_event_handler *, struct ib_event *);
629 struct list_head list;
632 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
634 (_ptr)->device = _device; \
635 (_ptr)->handler = _handler; \
636 INIT_LIST_HEAD(&(_ptr)->list); \
639 struct ib_global_route {
648 __be32 version_tclass_flow;
656 union rdma_network_hdr {
659 /* The IB spec states that if it's IPv4, the header
660 * is located in the last 20 bytes of the header.
663 struct iphdr roce4grh;
668 IB_MULTICAST_QPN = 0xffffff
671 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
672 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
679 IB_RATE_PORT_CURRENT = 0,
680 IB_RATE_2_5_GBPS = 2,
688 IB_RATE_120_GBPS = 10,
689 IB_RATE_14_GBPS = 11,
690 IB_RATE_56_GBPS = 12,
691 IB_RATE_112_GBPS = 13,
692 IB_RATE_168_GBPS = 14,
693 IB_RATE_25_GBPS = 15,
694 IB_RATE_100_GBPS = 16,
695 IB_RATE_200_GBPS = 17,
696 IB_RATE_300_GBPS = 18
700 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
701 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
702 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
703 * @rate: rate to convert.
705 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
708 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
709 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
710 * @rate: rate to convert.
712 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
716 * enum ib_mr_type - memory region type
717 * @IB_MR_TYPE_MEM_REG: memory region that is used for
718 * normal registration
719 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
720 * signature operations (data-integrity
722 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
723 * register any arbitrary sg lists (without
724 * the normal mr constraints - see
729 IB_MR_TYPE_SIGNATURE,
735 * IB_SIG_TYPE_NONE: Unprotected.
736 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
738 enum ib_signature_type {
744 * Signature T10-DIF block-guard types
745 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
746 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
748 enum ib_t10_dif_bg_type {
754 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
756 * @bg_type: T10-DIF block guard type (CRC|CSUM)
757 * @pi_interval: protection information interval.
758 * @bg: seed of guard computation.
759 * @app_tag: application tag of guard block
760 * @ref_tag: initial guard block reference tag.
761 * @ref_remap: Indicate wethear the reftag increments each block
762 * @app_escape: Indicate to skip block check if apptag=0xffff
763 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
764 * @apptag_check_mask: check bitmask of application tag.
766 struct ib_t10_dif_domain {
767 enum ib_t10_dif_bg_type bg_type;
775 u16 apptag_check_mask;
779 * struct ib_sig_domain - Parameters for signature domain
780 * @sig_type: specific signauture type
781 * @sig: union of all signature domain attributes that may
782 * be used to set domain layout.
784 struct ib_sig_domain {
785 enum ib_signature_type sig_type;
787 struct ib_t10_dif_domain dif;
792 * struct ib_sig_attrs - Parameters for signature handover operation
793 * @check_mask: bitmask for signature byte check (8 bytes)
794 * @mem: memory domain layout desciptor.
795 * @wire: wire domain layout desciptor.
797 struct ib_sig_attrs {
799 struct ib_sig_domain mem;
800 struct ib_sig_domain wire;
803 enum ib_sig_err_type {
810 * struct ib_sig_err - signature error descriptor
813 enum ib_sig_err_type err_type;
820 enum ib_mr_status_check {
821 IB_MR_CHECK_SIG_STATUS = 1,
825 * struct ib_mr_status - Memory region status container
827 * @fail_status: Bitmask of MR checks status. For each
828 * failed check a corresponding status bit is set.
829 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
832 struct ib_mr_status {
834 struct ib_sig_err sig_err;
838 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
840 * @mult: multiple to convert.
842 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
844 enum rdma_ah_attr_type {
845 RDMA_AH_ATTR_TYPE_IB,
846 RDMA_AH_ATTR_TYPE_ROCE,
847 RDMA_AH_ATTR_TYPE_OPA,
855 struct roce_ah_attr {
864 struct rdma_ah_attr {
865 struct ib_global_route grh;
870 enum rdma_ah_attr_type type;
872 struct ib_ah_attr ib;
873 struct roce_ah_attr roce;
874 struct opa_ah_attr opa;
882 IB_WC_LOC_EEC_OP_ERR,
887 IB_WC_LOC_ACCESS_ERR,
888 IB_WC_REM_INV_REQ_ERR,
889 IB_WC_REM_ACCESS_ERR,
892 IB_WC_RNR_RETRY_EXC_ERR,
893 IB_WC_LOC_RDD_VIOL_ERR,
894 IB_WC_REM_INV_RD_REQ_ERR,
897 IB_WC_INV_EEC_STATE_ERR,
899 IB_WC_RESP_TIMEOUT_ERR,
903 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
914 IB_WC_MASKED_COMP_SWAP,
915 IB_WC_MASKED_FETCH_ADD,
917 * Set value of IB_WC_RECV so consumers can test if a completion is a
918 * receive by testing (opcode & IB_WC_RECV).
921 IB_WC_RECV_RDMA_WITH_IMM
926 IB_WC_WITH_IMM = (1<<1),
927 IB_WC_WITH_INVALIDATE = (1<<2),
928 IB_WC_IP_CSUM_OK = (1<<3),
929 IB_WC_WITH_SMAC = (1<<4),
930 IB_WC_WITH_VLAN = (1<<5),
931 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
937 struct ib_cqe *wr_cqe;
939 enum ib_wc_status status;
940 enum ib_wc_opcode opcode;
954 u8 port_num; /* valid only for DR SMPs on switches */
960 enum ib_cq_notify_flags {
961 IB_CQ_SOLICITED = 1 << 0,
962 IB_CQ_NEXT_COMP = 1 << 1,
963 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
964 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
972 enum ib_srq_attr_mask {
973 IB_SRQ_MAX_WR = 1 << 0,
974 IB_SRQ_LIMIT = 1 << 1,
983 struct ib_srq_init_attr {
984 void (*event_handler)(struct ib_event *, void *);
986 struct ib_srq_attr attr;
987 enum ib_srq_type srq_type;
991 struct ib_xrcd *xrcd;
1002 u32 max_inline_data;
1005 * Maximum number of rdma_rw_ctx structures in flight at a time.
1006 * ib_create_qp() will calculate the right amount of neededed WRs
1007 * and MRs based on this.
1019 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1020 * here (and in that order) since the MAD layer uses them as
1021 * indices into a 2-entry table.
1030 IB_QPT_RAW_ETHERTYPE,
1031 IB_QPT_RAW_PACKET = 8,
1035 /* Reserve a range for qp types internal to the low level driver.
1036 * These qp types will not be visible at the IB core layer, so the
1037 * IB_QPT_MAX usages should not be affected in the core layer
1039 IB_QPT_RESERVED1 = 0x1000,
1051 enum ib_qp_create_flags {
1052 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1053 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1054 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1055 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1056 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1057 IB_QP_CREATE_NETIF_QP = 1 << 5,
1058 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1059 /* FREE = 1 << 7, */
1060 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1061 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1062 /* reserve bits 26-31 for low level drivers' internal use */
1063 IB_QP_CREATE_RESERVED_START = 1 << 26,
1064 IB_QP_CREATE_RESERVED_END = 1 << 31,
1068 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1069 * callback to destroy the passed in QP.
1072 struct ib_qp_init_attr {
1073 void (*event_handler)(struct ib_event *, void *);
1075 struct ib_cq *send_cq;
1076 struct ib_cq *recv_cq;
1078 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1079 struct ib_qp_cap cap;
1080 enum ib_sig_type sq_sig_type;
1081 enum ib_qp_type qp_type;
1082 enum ib_qp_create_flags create_flags;
1085 * Only needed for special QP types, or when using the RW API.
1088 struct ib_rwq_ind_table *rwq_ind_tbl;
1091 struct ib_qp_open_attr {
1092 void (*event_handler)(struct ib_event *, void *);
1095 enum ib_qp_type qp_type;
1098 enum ib_rnr_timeout {
1099 IB_RNR_TIMER_655_36 = 0,
1100 IB_RNR_TIMER_000_01 = 1,
1101 IB_RNR_TIMER_000_02 = 2,
1102 IB_RNR_TIMER_000_03 = 3,
1103 IB_RNR_TIMER_000_04 = 4,
1104 IB_RNR_TIMER_000_06 = 5,
1105 IB_RNR_TIMER_000_08 = 6,
1106 IB_RNR_TIMER_000_12 = 7,
1107 IB_RNR_TIMER_000_16 = 8,
1108 IB_RNR_TIMER_000_24 = 9,
1109 IB_RNR_TIMER_000_32 = 10,
1110 IB_RNR_TIMER_000_48 = 11,
1111 IB_RNR_TIMER_000_64 = 12,
1112 IB_RNR_TIMER_000_96 = 13,
1113 IB_RNR_TIMER_001_28 = 14,
1114 IB_RNR_TIMER_001_92 = 15,
1115 IB_RNR_TIMER_002_56 = 16,
1116 IB_RNR_TIMER_003_84 = 17,
1117 IB_RNR_TIMER_005_12 = 18,
1118 IB_RNR_TIMER_007_68 = 19,
1119 IB_RNR_TIMER_010_24 = 20,
1120 IB_RNR_TIMER_015_36 = 21,
1121 IB_RNR_TIMER_020_48 = 22,
1122 IB_RNR_TIMER_030_72 = 23,
1123 IB_RNR_TIMER_040_96 = 24,
1124 IB_RNR_TIMER_061_44 = 25,
1125 IB_RNR_TIMER_081_92 = 26,
1126 IB_RNR_TIMER_122_88 = 27,
1127 IB_RNR_TIMER_163_84 = 28,
1128 IB_RNR_TIMER_245_76 = 29,
1129 IB_RNR_TIMER_327_68 = 30,
1130 IB_RNR_TIMER_491_52 = 31
1133 enum ib_qp_attr_mask {
1135 IB_QP_CUR_STATE = (1<<1),
1136 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1137 IB_QP_ACCESS_FLAGS = (1<<3),
1138 IB_QP_PKEY_INDEX = (1<<4),
1139 IB_QP_PORT = (1<<5),
1140 IB_QP_QKEY = (1<<6),
1142 IB_QP_PATH_MTU = (1<<8),
1143 IB_QP_TIMEOUT = (1<<9),
1144 IB_QP_RETRY_CNT = (1<<10),
1145 IB_QP_RNR_RETRY = (1<<11),
1146 IB_QP_RQ_PSN = (1<<12),
1147 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1148 IB_QP_ALT_PATH = (1<<14),
1149 IB_QP_MIN_RNR_TIMER = (1<<15),
1150 IB_QP_SQ_PSN = (1<<16),
1151 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1152 IB_QP_PATH_MIG_STATE = (1<<18),
1153 IB_QP_CAP = (1<<19),
1154 IB_QP_DEST_QPN = (1<<20),
1155 IB_QP_RESERVED1 = (1<<21),
1156 IB_QP_RESERVED2 = (1<<22),
1157 IB_QP_RESERVED3 = (1<<23),
1158 IB_QP_RESERVED4 = (1<<24),
1159 IB_QP_RATE_LIMIT = (1<<25),
1184 enum ib_qp_state qp_state;
1185 enum ib_qp_state cur_qp_state;
1186 enum ib_mtu path_mtu;
1187 enum ib_mig_state path_mig_state;
1192 int qp_access_flags;
1193 struct ib_qp_cap cap;
1194 struct rdma_ah_attr ah_attr;
1195 struct rdma_ah_attr alt_ah_attr;
1198 u8 en_sqd_async_notify;
1201 u8 max_dest_rd_atomic;
1214 IB_WR_RDMA_WRITE_WITH_IMM,
1216 IB_WR_SEND_WITH_IMM,
1218 IB_WR_ATOMIC_CMP_AND_SWP,
1219 IB_WR_ATOMIC_FETCH_AND_ADD,
1221 IB_WR_SEND_WITH_INV,
1222 IB_WR_RDMA_READ_WITH_INV,
1225 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1226 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1228 /* reserve values for low level drivers' internal use.
1229 * These values will not be used at all in the ib core layer.
1231 IB_WR_RESERVED1 = 0xf0,
1243 enum ib_send_flags {
1245 IB_SEND_SIGNALED = (1<<1),
1246 IB_SEND_SOLICITED = (1<<2),
1247 IB_SEND_INLINE = (1<<3),
1248 IB_SEND_IP_CSUM = (1<<4),
1250 /* reserve bits 26-31 for low level drivers' internal use */
1251 IB_SEND_RESERVED_START = (1 << 26),
1252 IB_SEND_RESERVED_END = (1 << 31),
1262 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1266 struct ib_send_wr *next;
1269 struct ib_cqe *wr_cqe;
1271 struct ib_sge *sg_list;
1273 enum ib_wr_opcode opcode;
1277 u32 invalidate_rkey;
1282 struct ib_send_wr wr;
1287 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1289 return container_of(wr, struct ib_rdma_wr, wr);
1292 struct ib_atomic_wr {
1293 struct ib_send_wr wr;
1297 u64 compare_add_mask;
1302 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1304 return container_of(wr, struct ib_atomic_wr, wr);
1308 struct ib_send_wr wr;
1315 u16 pkey_index; /* valid for GSI only */
1316 u8 port_num; /* valid for DR SMPs on switch only */
1319 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1321 return container_of(wr, struct ib_ud_wr, wr);
1325 struct ib_send_wr wr;
1331 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1333 return container_of(wr, struct ib_reg_wr, wr);
1336 struct ib_sig_handover_wr {
1337 struct ib_send_wr wr;
1338 struct ib_sig_attrs *sig_attrs;
1339 struct ib_mr *sig_mr;
1341 struct ib_sge *prot;
1344 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1346 return container_of(wr, struct ib_sig_handover_wr, wr);
1350 struct ib_recv_wr *next;
1353 struct ib_cqe *wr_cqe;
1355 struct ib_sge *sg_list;
1359 enum ib_access_flags {
1360 IB_ACCESS_LOCAL_WRITE = 1,
1361 IB_ACCESS_REMOTE_WRITE = (1<<1),
1362 IB_ACCESS_REMOTE_READ = (1<<2),
1363 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1364 IB_ACCESS_MW_BIND = (1<<4),
1365 IB_ZERO_BASED = (1<<5),
1366 IB_ACCESS_ON_DEMAND = (1<<6),
1367 IB_ACCESS_HUGETLB = (1<<7),
1371 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1372 * are hidden here instead of a uapi header!
1374 enum ib_mr_rereg_flags {
1375 IB_MR_REREG_TRANS = 1,
1376 IB_MR_REREG_PD = (1<<1),
1377 IB_MR_REREG_ACCESS = (1<<2),
1378 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1381 struct ib_fmr_attr {
1389 enum rdma_remove_reason {
1390 /* Userspace requested uobject deletion. Call could fail */
1391 RDMA_REMOVE_DESTROY,
1392 /* Context deletion. This call should delete the actual object itself */
1394 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1395 RDMA_REMOVE_DRIVER_REMOVE,
1396 /* Context is being cleaned-up, but commit was just completed */
1397 RDMA_REMOVE_DURING_CLEANUP,
1400 struct ib_rdmacg_object {
1401 #ifdef CONFIG_CGROUP_RDMA
1402 struct rdma_cgroup *cg; /* owner rdma cgroup */
1406 struct ib_ucontext {
1407 struct ib_device *device;
1408 struct ib_uverbs_file *ufile;
1411 /* locking the uobjects_list */
1412 struct mutex uobjects_lock;
1413 struct list_head uobjects;
1414 /* protects cleanup process from other actions */
1415 struct rw_semaphore cleanup_rwsem;
1416 enum rdma_remove_reason cleanup_reason;
1419 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1420 struct rb_root umem_tree;
1422 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1423 * mmu notifiers registration.
1425 struct rw_semaphore umem_rwsem;
1426 void (*invalidate_range)(struct ib_umem *umem,
1427 unsigned long start, unsigned long end);
1429 struct mmu_notifier mn;
1430 atomic_t notifier_count;
1431 /* A list of umems that don't have private mmu notifier counters yet. */
1432 struct list_head no_private_counters;
1436 struct ib_rdmacg_object cg_obj;
1440 u64 user_handle; /* handle given to us by userspace */
1441 struct ib_ucontext *context; /* associated user context */
1442 void *object; /* containing object */
1443 struct list_head list; /* link to context's list */
1444 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1445 int id; /* index into kernel idr */
1447 atomic_t usecnt; /* protects exclusive access */
1448 struct rcu_head rcu; /* kfree_rcu() overhead */
1450 const struct uverbs_obj_type *type;
1453 struct ib_uobject_file {
1454 struct ib_uobject uobj;
1455 /* ufile contains the lock between context release and file close */
1456 struct ib_uverbs_file *ufile;
1460 const void __user *inbuf;
1461 void __user *outbuf;
1469 struct ib_device *device;
1470 struct ib_uobject *uobject;
1471 atomic_t usecnt; /* count all resources */
1473 u32 unsafe_global_rkey;
1476 * Implementation details of the RDMA core, don't use in drivers:
1478 struct ib_mr *__internal_mr;
1482 struct ib_device *device;
1483 atomic_t usecnt; /* count all exposed resources */
1484 struct inode *inode;
1486 struct mutex tgt_qp_mutex;
1487 struct list_head tgt_qp_list;
1491 struct ib_device *device;
1493 struct ib_uobject *uobject;
1494 enum rdma_ah_attr_type type;
1497 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1499 enum ib_poll_context {
1500 IB_POLL_DIRECT, /* caller context, no hw completions */
1501 IB_POLL_SOFTIRQ, /* poll from softirq context */
1502 IB_POLL_WORKQUEUE, /* poll from workqueue */
1506 struct ib_device *device;
1507 struct ib_uobject *uobject;
1508 ib_comp_handler comp_handler;
1509 void (*event_handler)(struct ib_event *, void *);
1512 atomic_t usecnt; /* count number of work queues */
1513 enum ib_poll_context poll_ctx;
1516 struct irq_poll iop;
1517 struct work_struct work;
1522 struct ib_device *device;
1524 struct ib_uobject *uobject;
1525 void (*event_handler)(struct ib_event *, void *);
1527 enum ib_srq_type srq_type;
1532 struct ib_xrcd *xrcd;
1539 enum ib_raw_packet_caps {
1540 /* Strip cvlan from incoming packet and report it in the matching work
1541 * completion is supported.
1543 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1544 /* Scatter FCS field of an incoming packet to host memory is supported.
1546 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1547 /* Checksum offloads are supported (for both send and receive). */
1548 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1562 struct ib_device *device;
1563 struct ib_uobject *uobject;
1565 void (*event_handler)(struct ib_event *, void *);
1569 enum ib_wq_state state;
1570 enum ib_wq_type wq_type;
1575 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1576 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1579 struct ib_wq_init_attr {
1581 enum ib_wq_type wq_type;
1585 void (*event_handler)(struct ib_event *, void *);
1586 u32 create_flags; /* Use enum ib_wq_flags */
1589 enum ib_wq_attr_mask {
1590 IB_WQ_STATE = 1 << 0,
1591 IB_WQ_CUR_STATE = 1 << 1,
1592 IB_WQ_FLAGS = 1 << 2,
1596 enum ib_wq_state wq_state;
1597 enum ib_wq_state curr_wq_state;
1598 u32 flags; /* Use enum ib_wq_flags */
1599 u32 flags_mask; /* Use enum ib_wq_flags */
1602 struct ib_rwq_ind_table {
1603 struct ib_device *device;
1604 struct ib_uobject *uobject;
1607 u32 log_ind_tbl_size;
1608 struct ib_wq **ind_tbl;
1611 struct ib_rwq_ind_table_init_attr {
1612 u32 log_ind_tbl_size;
1613 /* Each entry is a pointer to Receive Work Queue */
1614 struct ib_wq **ind_tbl;
1617 enum port_pkey_state {
1618 IB_PORT_PKEY_NOT_VALID = 0,
1619 IB_PORT_PKEY_VALID = 1,
1620 IB_PORT_PKEY_LISTED = 2,
1623 struct ib_qp_security;
1625 struct ib_port_pkey {
1626 enum port_pkey_state state;
1629 struct list_head qp_list;
1630 struct list_head to_error_list;
1631 struct ib_qp_security *sec;
1634 struct ib_ports_pkeys {
1635 struct ib_port_pkey main;
1636 struct ib_port_pkey alt;
1639 struct ib_qp_security {
1641 struct ib_device *dev;
1642 /* Hold this mutex when changing port and pkey settings. */
1644 struct ib_ports_pkeys *ports_pkeys;
1645 /* A list of all open shared QP handles. Required to enforce security
1646 * properly for all users of a shared QP.
1648 struct list_head shared_qp_list;
1651 atomic_t error_list_count;
1652 struct completion error_complete;
1653 int error_comps_pending;
1657 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1658 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1661 struct ib_device *device;
1663 struct ib_cq *send_cq;
1664 struct ib_cq *recv_cq;
1667 struct list_head rdma_mrs;
1668 struct list_head sig_mrs;
1670 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1671 struct list_head xrcd_list;
1673 /* count times opened, mcast attaches, flow attaches */
1675 struct list_head open_list;
1676 struct ib_qp *real_qp;
1677 struct ib_uobject *uobject;
1678 void (*event_handler)(struct ib_event *, void *);
1683 enum ib_qp_type qp_type;
1684 struct ib_rwq_ind_table *rwq_ind_tbl;
1685 struct ib_qp_security *qp_sec;
1690 struct ib_device *device;
1696 unsigned int page_size;
1699 struct ib_uobject *uobject; /* user */
1700 struct list_head qp_entry; /* FR */
1705 struct ib_device *device;
1707 struct ib_uobject *uobject;
1709 enum ib_mw_type type;
1713 struct ib_device *device;
1715 struct list_head list;
1720 /* Supported steering options */
1721 enum ib_flow_attr_type {
1722 /* steering according to rule specifications */
1723 IB_FLOW_ATTR_NORMAL = 0x0,
1724 /* default unicast and multicast rule -
1725 * receive all Eth traffic which isn't steered to any QP
1727 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1728 /* default multicast rule -
1729 * receive all Eth multicast traffic which isn't steered to any QP
1731 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1732 /* sniffer rule - receive all port traffic */
1733 IB_FLOW_ATTR_SNIFFER = 0x3
1736 /* Supported steering header types */
1737 enum ib_flow_spec_type {
1739 IB_FLOW_SPEC_ETH = 0x20,
1740 IB_FLOW_SPEC_IB = 0x22,
1742 IB_FLOW_SPEC_IPV4 = 0x30,
1743 IB_FLOW_SPEC_IPV6 = 0x31,
1745 IB_FLOW_SPEC_TCP = 0x40,
1746 IB_FLOW_SPEC_UDP = 0x41,
1747 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1748 IB_FLOW_SPEC_INNER = 0x100,
1750 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1751 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1753 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1754 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1756 /* Flow steering rule priority is set according to it's domain.
1757 * Lower domain value means higher priority.
1759 enum ib_flow_domain {
1760 IB_FLOW_DOMAIN_USER,
1761 IB_FLOW_DOMAIN_ETHTOOL,
1764 IB_FLOW_DOMAIN_NUM /* Must be last */
1767 enum ib_flow_flags {
1768 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1769 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1772 struct ib_flow_eth_filter {
1781 struct ib_flow_spec_eth {
1784 struct ib_flow_eth_filter val;
1785 struct ib_flow_eth_filter mask;
1788 struct ib_flow_ib_filter {
1795 struct ib_flow_spec_ib {
1798 struct ib_flow_ib_filter val;
1799 struct ib_flow_ib_filter mask;
1802 /* IPv4 header flags */
1803 enum ib_ipv4_flags {
1804 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1805 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1806 last have this flag set */
1809 struct ib_flow_ipv4_filter {
1820 struct ib_flow_spec_ipv4 {
1823 struct ib_flow_ipv4_filter val;
1824 struct ib_flow_ipv4_filter mask;
1827 struct ib_flow_ipv6_filter {
1838 struct ib_flow_spec_ipv6 {
1841 struct ib_flow_ipv6_filter val;
1842 struct ib_flow_ipv6_filter mask;
1845 struct ib_flow_tcp_udp_filter {
1852 struct ib_flow_spec_tcp_udp {
1855 struct ib_flow_tcp_udp_filter val;
1856 struct ib_flow_tcp_udp_filter mask;
1859 struct ib_flow_tunnel_filter {
1864 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1865 * the tunnel_id from val has the vni value
1867 struct ib_flow_spec_tunnel {
1870 struct ib_flow_tunnel_filter val;
1871 struct ib_flow_tunnel_filter mask;
1874 struct ib_flow_spec_action_tag {
1875 enum ib_flow_spec_type type;
1880 struct ib_flow_spec_action_drop {
1881 enum ib_flow_spec_type type;
1885 union ib_flow_spec {
1890 struct ib_flow_spec_eth eth;
1891 struct ib_flow_spec_ib ib;
1892 struct ib_flow_spec_ipv4 ipv4;
1893 struct ib_flow_spec_tcp_udp tcp_udp;
1894 struct ib_flow_spec_ipv6 ipv6;
1895 struct ib_flow_spec_tunnel tunnel;
1896 struct ib_flow_spec_action_tag flow_tag;
1897 struct ib_flow_spec_action_drop drop;
1900 struct ib_flow_attr {
1901 enum ib_flow_attr_type type;
1907 /* Following are the optional layers according to user request
1908 * struct ib_flow_spec_xxx
1909 * struct ib_flow_spec_yyy
1915 struct ib_uobject *uobject;
1921 enum ib_process_mad_flags {
1922 IB_MAD_IGNORE_MKEY = 1,
1923 IB_MAD_IGNORE_BKEY = 2,
1924 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1927 enum ib_mad_result {
1928 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1929 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1930 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1931 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1934 struct ib_port_cache {
1936 struct ib_pkey_cache *pkey;
1937 struct ib_gid_table *gid;
1939 enum ib_port_state port_state;
1944 struct ib_event_handler event_handler;
1945 struct ib_port_cache *ports;
1950 struct ib_port_immutable {
1957 /* rdma netdev type - specifies protocol type */
1958 enum rdma_netdev_t {
1959 RDMA_NETDEV_OPA_VNIC,
1964 * struct rdma_netdev - rdma netdev
1965 * For cases where netstack interfacing is required.
1967 struct rdma_netdev {
1969 struct ib_device *hca;
1972 /* cleanup function must be specified */
1973 void (*free_rdma_netdev)(struct net_device *netdev);
1975 /* control functions */
1976 void (*set_id)(struct net_device *netdev, int id);
1978 int (*send)(struct net_device *dev, struct sk_buff *skb,
1979 struct ib_ah *address, u32 dqpn);
1981 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
1982 union ib_gid *gid, u16 mlid,
1983 int set_qkey, u32 qkey);
1984 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
1985 union ib_gid *gid, u16 mlid);
1988 struct ib_port_pkey_list {
1989 /* Lock to hold while modifying the list. */
1990 spinlock_t list_lock;
1991 struct list_head pkey_list;
1995 /* Do not access @dma_device directly from ULP nor from HW drivers. */
1996 struct device *dma_device;
1998 char name[IB_DEVICE_NAME_MAX];
2000 struct list_head event_handler_list;
2001 spinlock_t event_handler_lock;
2003 spinlock_t client_data_lock;
2004 struct list_head core_list;
2005 /* Access to the client_data_list is protected by the client_data_lock
2006 * spinlock and the lists_rwsem read-write semaphore */
2007 struct list_head client_data_list;
2009 struct ib_cache cache;
2011 * port_immutable is indexed by port number
2013 struct ib_port_immutable *port_immutable;
2015 int num_comp_vectors;
2017 struct ib_port_pkey_list *port_pkey_list;
2019 struct iw_cm_verbs *iwcm;
2022 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2023 * driver initialized data. The struct is kfree()'ed by the sysfs
2024 * core when the device is removed. A lifespan of -1 in the return
2025 * struct tells the core to set a default lifespan.
2027 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2030 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2031 * @index - The index in the value array we wish to have updated, or
2032 * num_counters if we want all stats updated
2034 * < 0 - Error, no counters updated
2035 * index - Updated the single counter pointed to by index
2036 * num_counters - Updated all counters (will reset the timestamp
2037 * and prevent further calls for lifespan milliseconds)
2038 * Drivers are allowed to update all counters in leiu of just the
2039 * one given in index at their option
2041 int (*get_hw_stats)(struct ib_device *device,
2042 struct rdma_hw_stats *stats,
2043 u8 port, int index);
2044 int (*query_device)(struct ib_device *device,
2045 struct ib_device_attr *device_attr,
2046 struct ib_udata *udata);
2047 int (*query_port)(struct ib_device *device,
2049 struct ib_port_attr *port_attr);
2050 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2052 /* When calling get_netdev, the HW vendor's driver should return the
2053 * net device of device @device at port @port_num or NULL if such
2054 * a net device doesn't exist. The vendor driver should call dev_hold
2055 * on this net device. The HW vendor's device driver must guarantee
2056 * that this function returns NULL before the net device reaches
2057 * NETDEV_UNREGISTER_FINAL state.
2059 struct net_device *(*get_netdev)(struct ib_device *device,
2061 int (*query_gid)(struct ib_device *device,
2062 u8 port_num, int index,
2064 /* When calling add_gid, the HW vendor's driver should
2065 * add the gid of device @device at gid index @index of
2066 * port @port_num to be @gid. Meta-info of that gid (for example,
2067 * the network device related to this gid is available
2068 * at @attr. @context allows the HW vendor driver to store extra
2069 * information together with a GID entry. The HW vendor may allocate
2070 * memory to contain this information and store it in @context when a
2071 * new GID entry is written to. Params are consistent until the next
2072 * call of add_gid or delete_gid. The function should return 0 on
2073 * success or error otherwise. The function could be called
2074 * concurrently for different ports. This function is only called
2075 * when roce_gid_table is used.
2077 int (*add_gid)(struct ib_device *device,
2080 const union ib_gid *gid,
2081 const struct ib_gid_attr *attr,
2083 /* When calling del_gid, the HW vendor's driver should delete the
2084 * gid of device @device at gid index @index of port @port_num.
2085 * Upon the deletion of a GID entry, the HW vendor must free any
2086 * allocated memory. The caller will clear @context afterwards.
2087 * This function is only called when roce_gid_table is used.
2089 int (*del_gid)(struct ib_device *device,
2093 int (*query_pkey)(struct ib_device *device,
2094 u8 port_num, u16 index, u16 *pkey);
2095 int (*modify_device)(struct ib_device *device,
2096 int device_modify_mask,
2097 struct ib_device_modify *device_modify);
2098 int (*modify_port)(struct ib_device *device,
2099 u8 port_num, int port_modify_mask,
2100 struct ib_port_modify *port_modify);
2101 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2102 struct ib_udata *udata);
2103 int (*dealloc_ucontext)(struct ib_ucontext *context);
2104 int (*mmap)(struct ib_ucontext *context,
2105 struct vm_area_struct *vma);
2106 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2107 struct ib_ucontext *context,
2108 struct ib_udata *udata);
2109 int (*dealloc_pd)(struct ib_pd *pd);
2110 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2111 struct rdma_ah_attr *ah_attr,
2112 struct ib_udata *udata);
2113 int (*modify_ah)(struct ib_ah *ah,
2114 struct rdma_ah_attr *ah_attr);
2115 int (*query_ah)(struct ib_ah *ah,
2116 struct rdma_ah_attr *ah_attr);
2117 int (*destroy_ah)(struct ib_ah *ah);
2118 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2119 struct ib_srq_init_attr *srq_init_attr,
2120 struct ib_udata *udata);
2121 int (*modify_srq)(struct ib_srq *srq,
2122 struct ib_srq_attr *srq_attr,
2123 enum ib_srq_attr_mask srq_attr_mask,
2124 struct ib_udata *udata);
2125 int (*query_srq)(struct ib_srq *srq,
2126 struct ib_srq_attr *srq_attr);
2127 int (*destroy_srq)(struct ib_srq *srq);
2128 int (*post_srq_recv)(struct ib_srq *srq,
2129 struct ib_recv_wr *recv_wr,
2130 struct ib_recv_wr **bad_recv_wr);
2131 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2132 struct ib_qp_init_attr *qp_init_attr,
2133 struct ib_udata *udata);
2134 int (*modify_qp)(struct ib_qp *qp,
2135 struct ib_qp_attr *qp_attr,
2137 struct ib_udata *udata);
2138 int (*query_qp)(struct ib_qp *qp,
2139 struct ib_qp_attr *qp_attr,
2141 struct ib_qp_init_attr *qp_init_attr);
2142 int (*destroy_qp)(struct ib_qp *qp);
2143 int (*post_send)(struct ib_qp *qp,
2144 struct ib_send_wr *send_wr,
2145 struct ib_send_wr **bad_send_wr);
2146 int (*post_recv)(struct ib_qp *qp,
2147 struct ib_recv_wr *recv_wr,
2148 struct ib_recv_wr **bad_recv_wr);
2149 struct ib_cq * (*create_cq)(struct ib_device *device,
2150 const struct ib_cq_init_attr *attr,
2151 struct ib_ucontext *context,
2152 struct ib_udata *udata);
2153 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2155 int (*destroy_cq)(struct ib_cq *cq);
2156 int (*resize_cq)(struct ib_cq *cq, int cqe,
2157 struct ib_udata *udata);
2158 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2160 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2161 int (*req_notify_cq)(struct ib_cq *cq,
2162 enum ib_cq_notify_flags flags);
2163 int (*req_ncomp_notif)(struct ib_cq *cq,
2165 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2166 int mr_access_flags);
2167 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2168 u64 start, u64 length,
2170 int mr_access_flags,
2171 struct ib_udata *udata);
2172 int (*rereg_user_mr)(struct ib_mr *mr,
2174 u64 start, u64 length,
2176 int mr_access_flags,
2178 struct ib_udata *udata);
2179 int (*dereg_mr)(struct ib_mr *mr);
2180 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2181 enum ib_mr_type mr_type,
2183 int (*map_mr_sg)(struct ib_mr *mr,
2184 struct scatterlist *sg,
2186 unsigned int *sg_offset);
2187 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2188 enum ib_mw_type type,
2189 struct ib_udata *udata);
2190 int (*dealloc_mw)(struct ib_mw *mw);
2191 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2192 int mr_access_flags,
2193 struct ib_fmr_attr *fmr_attr);
2194 int (*map_phys_fmr)(struct ib_fmr *fmr,
2195 u64 *page_list, int list_len,
2197 int (*unmap_fmr)(struct list_head *fmr_list);
2198 int (*dealloc_fmr)(struct ib_fmr *fmr);
2199 int (*attach_mcast)(struct ib_qp *qp,
2202 int (*detach_mcast)(struct ib_qp *qp,
2205 int (*process_mad)(struct ib_device *device,
2206 int process_mad_flags,
2208 const struct ib_wc *in_wc,
2209 const struct ib_grh *in_grh,
2210 const struct ib_mad_hdr *in_mad,
2212 struct ib_mad_hdr *out_mad,
2213 size_t *out_mad_size,
2214 u16 *out_mad_pkey_index);
2215 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2216 struct ib_ucontext *ucontext,
2217 struct ib_udata *udata);
2218 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2219 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2223 int (*destroy_flow)(struct ib_flow *flow_id);
2224 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2225 struct ib_mr_status *mr_status);
2226 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2227 void (*drain_rq)(struct ib_qp *qp);
2228 void (*drain_sq)(struct ib_qp *qp);
2229 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2231 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2232 struct ifla_vf_info *ivf);
2233 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2234 struct ifla_vf_stats *stats);
2235 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2237 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2238 struct ib_wq_init_attr *init_attr,
2239 struct ib_udata *udata);
2240 int (*destroy_wq)(struct ib_wq *wq);
2241 int (*modify_wq)(struct ib_wq *wq,
2242 struct ib_wq_attr *attr,
2244 struct ib_udata *udata);
2245 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2246 struct ib_rwq_ind_table_init_attr *init_attr,
2247 struct ib_udata *udata);
2248 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2250 * rdma netdev operation
2252 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2253 * doesn't support the specified rdma netdev type.
2255 struct net_device *(*alloc_rdma_netdev)(
2256 struct ib_device *device,
2258 enum rdma_netdev_t type,
2260 unsigned char name_assign_type,
2261 void (*setup)(struct net_device *));
2263 struct module *owner;
2265 struct kobject *ports_parent;
2266 struct list_head port_list;
2269 IB_DEV_UNINITIALIZED,
2275 u64 uverbs_cmd_mask;
2276 u64 uverbs_ex_cmd_mask;
2278 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2284 struct ib_device_attr attrs;
2285 struct attribute_group *hw_stats_ag;
2286 struct rdma_hw_stats *hw_stats;
2288 #ifdef CONFIG_CGROUP_RDMA
2289 struct rdmacg_device cg_device;
2293 * The following mandatory functions are used only at device
2294 * registration. Keep functions such as these at the end of this
2295 * structure to avoid cache line misses when accessing struct ib_device
2298 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2299 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2304 void (*add) (struct ib_device *);
2305 void (*remove)(struct ib_device *, void *client_data);
2307 /* Returns the net_dev belonging to this ib_client and matching the
2309 * @dev: An RDMA device that the net_dev use for communication.
2310 * @port: A physical port number on the RDMA device.
2311 * @pkey: P_Key that the net_dev uses if applicable.
2312 * @gid: A GID that the net_dev uses to communicate.
2313 * @addr: An IP address the net_dev is configured with.
2314 * @client_data: The device's client data set by ib_set_client_data().
2316 * An ib_client that implements a net_dev on top of RDMA devices
2317 * (such as IP over IB) should implement this callback, allowing the
2318 * rdma_cm module to find the right net_dev for a given request.
2320 * The caller is responsible for calling dev_put on the returned
2322 struct net_device *(*get_net_dev_by_params)(
2323 struct ib_device *dev,
2326 const union ib_gid *gid,
2327 const struct sockaddr *addr,
2329 struct list_head list;
2332 struct ib_device *ib_alloc_device(size_t size);
2333 void ib_dealloc_device(struct ib_device *device);
2335 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2337 int ib_register_device(struct ib_device *device,
2338 int (*port_callback)(struct ib_device *,
2339 u8, struct kobject *));
2340 void ib_unregister_device(struct ib_device *device);
2342 int ib_register_client (struct ib_client *client);
2343 void ib_unregister_client(struct ib_client *client);
2345 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2346 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2349 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2351 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2354 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2356 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2359 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2363 const void __user *p = udata->inbuf + offset;
2367 if (len > USHRT_MAX)
2370 buf = memdup_user(p, len);
2374 ret = !memchr_inv(buf, 0, len);
2380 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2381 * contains all required attributes and no attributes not allowed for
2382 * the given QP state transition.
2383 * @cur_state: Current QP state
2384 * @next_state: Next QP state
2386 * @mask: Mask of supplied QP attributes
2387 * @ll : link layer of port
2389 * This function is a helper function that a low-level driver's
2390 * modify_qp method can use to validate the consumer's input. It
2391 * checks that cur_state and next_state are valid QP states, that a
2392 * transition from cur_state to next_state is allowed by the IB spec,
2393 * and that the attribute mask supplied is allowed for the transition.
2395 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2396 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2397 enum rdma_link_layer ll);
2399 int ib_register_event_handler (struct ib_event_handler *event_handler);
2400 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2401 void ib_dispatch_event(struct ib_event *event);
2403 int ib_query_port(struct ib_device *device,
2404 u8 port_num, struct ib_port_attr *port_attr);
2406 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2410 * rdma_cap_ib_switch - Check if the device is IB switch
2411 * @device: Device to check
2413 * Device driver is responsible for setting is_switch bit on
2414 * in ib_device structure at init time.
2416 * Return: true if the device is IB switch.
2418 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2420 return device->is_switch;
2424 * rdma_start_port - Return the first valid port number for the device
2427 * @device: Device to be checked
2429 * Return start port number
2431 static inline u8 rdma_start_port(const struct ib_device *device)
2433 return rdma_cap_ib_switch(device) ? 0 : 1;
2437 * rdma_end_port - Return the last valid port number for the device
2440 * @device: Device to be checked
2442 * Return last port number
2444 static inline u8 rdma_end_port(const struct ib_device *device)
2446 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2449 static inline int rdma_is_port_valid(const struct ib_device *device,
2452 return (port >= rdma_start_port(device) &&
2453 port <= rdma_end_port(device));
2456 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2458 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2461 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2463 return device->port_immutable[port_num].core_cap_flags &
2464 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2467 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2469 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2472 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2474 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2477 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2479 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2482 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2484 return rdma_protocol_ib(device, port_num) ||
2485 rdma_protocol_roce(device, port_num);
2488 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2490 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2493 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2495 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2499 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2500 * Management Datagrams.
2501 * @device: Device to check
2502 * @port_num: Port number to check
2504 * Management Datagrams (MAD) are a required part of the InfiniBand
2505 * specification and are supported on all InfiniBand devices. A slightly
2506 * extended version are also supported on OPA interfaces.
2508 * Return: true if the port supports sending/receiving of MAD packets.
2510 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2512 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2516 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2517 * Management Datagrams.
2518 * @device: Device to check
2519 * @port_num: Port number to check
2521 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2522 * datagrams with their own versions. These OPA MADs share many but not all of
2523 * the characteristics of InfiniBand MADs.
2525 * OPA MADs differ in the following ways:
2527 * 1) MADs are variable size up to 2K
2528 * IBTA defined MADs remain fixed at 256 bytes
2529 * 2) OPA SMPs must carry valid PKeys
2530 * 3) OPA SMP packets are a different format
2532 * Return: true if the port supports OPA MAD packet formats.
2534 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2536 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2537 == RDMA_CORE_CAP_OPA_MAD;
2541 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2542 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2543 * @device: Device to check
2544 * @port_num: Port number to check
2546 * Each InfiniBand node is required to provide a Subnet Management Agent
2547 * that the subnet manager can access. Prior to the fabric being fully
2548 * configured by the subnet manager, the SMA is accessed via a well known
2549 * interface called the Subnet Management Interface (SMI). This interface
2550 * uses directed route packets to communicate with the SM to get around the
2551 * chicken and egg problem of the SM needing to know what's on the fabric
2552 * in order to configure the fabric, and needing to configure the fabric in
2553 * order to send packets to the devices on the fabric. These directed
2554 * route packets do not need the fabric fully configured in order to reach
2555 * their destination. The SMI is the only method allowed to send
2556 * directed route packets on an InfiniBand fabric.
2558 * Return: true if the port provides an SMI.
2560 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2562 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2566 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2567 * Communication Manager.
2568 * @device: Device to check
2569 * @port_num: Port number to check
2571 * The InfiniBand Communication Manager is one of many pre-defined General
2572 * Service Agents (GSA) that are accessed via the General Service
2573 * Interface (GSI). It's role is to facilitate establishment of connections
2574 * between nodes as well as other management related tasks for established
2577 * Return: true if the port supports an IB CM (this does not guarantee that
2578 * a CM is actually running however).
2580 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2582 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2586 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2587 * Communication Manager.
2588 * @device: Device to check
2589 * @port_num: Port number to check
2591 * Similar to above, but specific to iWARP connections which have a different
2592 * managment protocol than InfiniBand.
2594 * Return: true if the port supports an iWARP CM (this does not guarantee that
2595 * a CM is actually running however).
2597 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2599 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2603 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2604 * Subnet Administration.
2605 * @device: Device to check
2606 * @port_num: Port number to check
2608 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2609 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2610 * fabrics, devices should resolve routes to other hosts by contacting the
2611 * SA to query the proper route.
2613 * Return: true if the port should act as a client to the fabric Subnet
2614 * Administration interface. This does not imply that the SA service is
2617 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2619 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2623 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2625 * @device: Device to check
2626 * @port_num: Port number to check
2628 * InfiniBand multicast registration is more complex than normal IPv4 or
2629 * IPv6 multicast registration. Each Host Channel Adapter must register
2630 * with the Subnet Manager when it wishes to join a multicast group. It
2631 * should do so only once regardless of how many queue pairs it subscribes
2632 * to this group. And it should leave the group only after all queue pairs
2633 * attached to the group have been detached.
2635 * Return: true if the port must undertake the additional adminstrative
2636 * overhead of registering/unregistering with the SM and tracking of the
2637 * total number of queue pairs attached to the multicast group.
2639 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2641 return rdma_cap_ib_sa(device, port_num);
2645 * rdma_cap_af_ib - Check if the port of device has the capability
2646 * Native Infiniband Address.
2647 * @device: Device to check
2648 * @port_num: Port number to check
2650 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2651 * GID. RoCE uses a different mechanism, but still generates a GID via
2652 * a prescribed mechanism and port specific data.
2654 * Return: true if the port uses a GID address to identify devices on the
2657 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2659 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2663 * rdma_cap_eth_ah - Check if the port of device has the capability
2664 * Ethernet Address Handle.
2665 * @device: Device to check
2666 * @port_num: Port number to check
2668 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2669 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2670 * port. Normally, packet headers are generated by the sending host
2671 * adapter, but when sending connectionless datagrams, we must manually
2672 * inject the proper headers for the fabric we are communicating over.
2674 * Return: true if we are running as a RoCE port and must force the
2675 * addition of a Global Route Header built from our Ethernet Address
2676 * Handle into our header list for connectionless packets.
2678 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2680 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2684 * rdma_cap_opa_ah - Check if the port of device supports
2685 * OPA Address handles
2686 * @device: Device to check
2687 * @port_num: Port number to check
2689 * Return: true if we are running on an OPA device which supports
2690 * the extended OPA addressing.
2692 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2694 return (device->port_immutable[port_num].core_cap_flags &
2695 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2699 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2702 * @port_num: Port number
2704 * This MAD size includes the MAD headers and MAD payload. No other headers
2707 * Return the max MAD size required by the Port. Will return 0 if the port
2708 * does not support MADs
2710 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2712 return device->port_immutable[port_num].max_mad_size;
2716 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2717 * @device: Device to check
2718 * @port_num: Port number to check
2720 * RoCE GID table mechanism manages the various GIDs for a device.
2722 * NOTE: if allocating the port's GID table has failed, this call will still
2723 * return true, but any RoCE GID table API will fail.
2725 * Return: true if the port uses RoCE GID table mechanism in order to manage
2728 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2731 return rdma_protocol_roce(device, port_num) &&
2732 device->add_gid && device->del_gid;
2736 * Check if the device supports READ W/ INVALIDATE.
2738 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2741 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2742 * has support for it yet.
2744 return rdma_protocol_iwarp(dev, port_num);
2747 int ib_query_gid(struct ib_device *device,
2748 u8 port_num, int index, union ib_gid *gid,
2749 struct ib_gid_attr *attr);
2751 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2753 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2754 struct ifla_vf_info *info);
2755 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2756 struct ifla_vf_stats *stats);
2757 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2760 int ib_query_pkey(struct ib_device *device,
2761 u8 port_num, u16 index, u16 *pkey);
2763 int ib_modify_device(struct ib_device *device,
2764 int device_modify_mask,
2765 struct ib_device_modify *device_modify);
2767 int ib_modify_port(struct ib_device *device,
2768 u8 port_num, int port_modify_mask,
2769 struct ib_port_modify *port_modify);
2771 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2772 enum ib_gid_type gid_type, struct net_device *ndev,
2773 u8 *port_num, u16 *index);
2775 int ib_find_pkey(struct ib_device *device,
2776 u8 port_num, u16 pkey, u16 *index);
2780 * Create a memory registration for all memory in the system and place
2781 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2782 * ULPs to avoid the overhead of dynamic MRs.
2784 * This flag is generally considered unsafe and must only be used in
2785 * extremly trusted environments. Every use of it will log a warning
2786 * in the kernel log.
2788 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2791 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2792 const char *caller);
2793 #define ib_alloc_pd(device, flags) \
2794 __ib_alloc_pd((device), (flags), __func__)
2795 void ib_dealloc_pd(struct ib_pd *pd);
2798 * rdma_create_ah - Creates an address handle for the given address vector.
2799 * @pd: The protection domain associated with the address handle.
2800 * @ah_attr: The attributes of the address vector.
2802 * The address handle is used to reference a local or global destination
2803 * in all UD QP post sends.
2805 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2808 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2810 * @hdr: the L3 header to parse
2811 * @net_type: type of header to parse
2812 * @sgid: place to store source gid
2813 * @dgid: place to store destination gid
2815 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2816 enum rdma_network_type net_type,
2817 union ib_gid *sgid, union ib_gid *dgid);
2820 * ib_get_rdma_header_version - Get the header version
2821 * @hdr: the L3 header to parse
2823 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2826 * ib_init_ah_from_wc - Initializes address handle attributes from a
2828 * @device: Device on which the received message arrived.
2829 * @port_num: Port on which the received message arrived.
2830 * @wc: Work completion associated with the received message.
2831 * @grh: References the received global route header. This parameter is
2832 * ignored unless the work completion indicates that the GRH is valid.
2833 * @ah_attr: Returned attributes that can be used when creating an address
2834 * handle for replying to the message.
2836 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2837 const struct ib_wc *wc, const struct ib_grh *grh,
2838 struct rdma_ah_attr *ah_attr);
2841 * ib_create_ah_from_wc - Creates an address handle associated with the
2842 * sender of the specified work completion.
2843 * @pd: The protection domain associated with the address handle.
2844 * @wc: Work completion information associated with a received message.
2845 * @grh: References the received global route header. This parameter is
2846 * ignored unless the work completion indicates that the GRH is valid.
2847 * @port_num: The outbound port number to associate with the address.
2849 * The address handle is used to reference a local or global destination
2850 * in all UD QP post sends.
2852 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2853 const struct ib_grh *grh, u8 port_num);
2856 * rdma_modify_ah - Modifies the address vector associated with an address
2858 * @ah: The address handle to modify.
2859 * @ah_attr: The new address vector attributes to associate with the
2862 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2865 * rdma_query_ah - Queries the address vector associated with an address
2867 * @ah: The address handle to query.
2868 * @ah_attr: The address vector attributes associated with the address
2871 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2874 * rdma_destroy_ah - Destroys an address handle.
2875 * @ah: The address handle to destroy.
2877 int rdma_destroy_ah(struct ib_ah *ah);
2880 * ib_create_srq - Creates a SRQ associated with the specified protection
2882 * @pd: The protection domain associated with the SRQ.
2883 * @srq_init_attr: A list of initial attributes required to create the
2884 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2885 * the actual capabilities of the created SRQ.
2887 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2888 * requested size of the SRQ, and set to the actual values allocated
2889 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2890 * will always be at least as large as the requested values.
2892 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2893 struct ib_srq_init_attr *srq_init_attr);
2896 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2897 * @srq: The SRQ to modify.
2898 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2899 * the current values of selected SRQ attributes are returned.
2900 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2901 * are being modified.
2903 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2904 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2905 * the number of receives queued drops below the limit.
2907 int ib_modify_srq(struct ib_srq *srq,
2908 struct ib_srq_attr *srq_attr,
2909 enum ib_srq_attr_mask srq_attr_mask);
2912 * ib_query_srq - Returns the attribute list and current values for the
2914 * @srq: The SRQ to query.
2915 * @srq_attr: The attributes of the specified SRQ.
2917 int ib_query_srq(struct ib_srq *srq,
2918 struct ib_srq_attr *srq_attr);
2921 * ib_destroy_srq - Destroys the specified SRQ.
2922 * @srq: The SRQ to destroy.
2924 int ib_destroy_srq(struct ib_srq *srq);
2927 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2928 * @srq: The SRQ to post the work request on.
2929 * @recv_wr: A list of work requests to post on the receive queue.
2930 * @bad_recv_wr: On an immediate failure, this parameter will reference
2931 * the work request that failed to be posted on the QP.
2933 static inline int ib_post_srq_recv(struct ib_srq *srq,
2934 struct ib_recv_wr *recv_wr,
2935 struct ib_recv_wr **bad_recv_wr)
2937 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2941 * ib_create_qp - Creates a QP associated with the specified protection
2943 * @pd: The protection domain associated with the QP.
2944 * @qp_init_attr: A list of initial attributes required to create the
2945 * QP. If QP creation succeeds, then the attributes are updated to
2946 * the actual capabilities of the created QP.
2948 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2949 struct ib_qp_init_attr *qp_init_attr);
2952 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
2953 * @qp: The QP to modify.
2954 * @attr: On input, specifies the QP attributes to modify. On output,
2955 * the current values of selected QP attributes are returned.
2956 * @attr_mask: A bit-mask used to specify which attributes of the QP
2957 * are being modified.
2958 * @udata: pointer to user's input output buffer information
2959 * are being modified.
2960 * It returns 0 on success and returns appropriate error code on error.
2962 int ib_modify_qp_with_udata(struct ib_qp *qp,
2963 struct ib_qp_attr *attr,
2965 struct ib_udata *udata);
2968 * ib_modify_qp - Modifies the attributes for the specified QP and then
2969 * transitions the QP to the given state.
2970 * @qp: The QP to modify.
2971 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2972 * the current values of selected QP attributes are returned.
2973 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2974 * are being modified.
2976 int ib_modify_qp(struct ib_qp *qp,
2977 struct ib_qp_attr *qp_attr,
2981 * ib_query_qp - Returns the attribute list and current values for the
2983 * @qp: The QP to query.
2984 * @qp_attr: The attributes of the specified QP.
2985 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2986 * @qp_init_attr: Additional attributes of the selected QP.
2988 * The qp_attr_mask may be used to limit the query to gathering only the
2989 * selected attributes.
2991 int ib_query_qp(struct ib_qp *qp,
2992 struct ib_qp_attr *qp_attr,
2994 struct ib_qp_init_attr *qp_init_attr);
2997 * ib_destroy_qp - Destroys the specified QP.
2998 * @qp: The QP to destroy.
3000 int ib_destroy_qp(struct ib_qp *qp);
3003 * ib_open_qp - Obtain a reference to an existing sharable QP.
3004 * @xrcd - XRC domain
3005 * @qp_open_attr: Attributes identifying the QP to open.
3007 * Returns a reference to a sharable QP.
3009 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3010 struct ib_qp_open_attr *qp_open_attr);
3013 * ib_close_qp - Release an external reference to a QP.
3014 * @qp: The QP handle to release
3016 * The opened QP handle is released by the caller. The underlying
3017 * shared QP is not destroyed until all internal references are released.
3019 int ib_close_qp(struct ib_qp *qp);
3022 * ib_post_send - Posts a list of work requests to the send queue of
3024 * @qp: The QP to post the work request on.
3025 * @send_wr: A list of work requests to post on the send queue.
3026 * @bad_send_wr: On an immediate failure, this parameter will reference
3027 * the work request that failed to be posted on the QP.
3029 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3030 * error is returned, the QP state shall not be affected,
3031 * ib_post_send() will return an immediate error after queueing any
3032 * earlier work requests in the list.
3034 static inline int ib_post_send(struct ib_qp *qp,
3035 struct ib_send_wr *send_wr,
3036 struct ib_send_wr **bad_send_wr)
3038 return qp->device->post_send(qp, send_wr, bad_send_wr);
3042 * ib_post_recv - Posts a list of work requests to the receive queue of
3044 * @qp: The QP to post the work request on.
3045 * @recv_wr: A list of work requests to post on the receive queue.
3046 * @bad_recv_wr: On an immediate failure, this parameter will reference
3047 * the work request that failed to be posted on the QP.
3049 static inline int ib_post_recv(struct ib_qp *qp,
3050 struct ib_recv_wr *recv_wr,
3051 struct ib_recv_wr **bad_recv_wr)
3053 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3056 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3057 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
3058 void ib_free_cq(struct ib_cq *cq);
3059 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3062 * ib_create_cq - Creates a CQ on the specified device.
3063 * @device: The device on which to create the CQ.
3064 * @comp_handler: A user-specified callback that is invoked when a
3065 * completion event occurs on the CQ.
3066 * @event_handler: A user-specified callback that is invoked when an
3067 * asynchronous event not associated with a completion occurs on the CQ.
3068 * @cq_context: Context associated with the CQ returned to the user via
3069 * the associated completion and event handlers.
3070 * @cq_attr: The attributes the CQ should be created upon.
3072 * Users can examine the cq structure to determine the actual CQ size.
3074 struct ib_cq *ib_create_cq(struct ib_device *device,
3075 ib_comp_handler comp_handler,
3076 void (*event_handler)(struct ib_event *, void *),
3078 const struct ib_cq_init_attr *cq_attr);
3081 * ib_resize_cq - Modifies the capacity of the CQ.
3082 * @cq: The CQ to resize.
3083 * @cqe: The minimum size of the CQ.
3085 * Users can examine the cq structure to determine the actual CQ size.
3087 int ib_resize_cq(struct ib_cq *cq, int cqe);
3090 * ib_modify_cq - Modifies moderation params of the CQ
3091 * @cq: The CQ to modify.
3092 * @cq_count: number of CQEs that will trigger an event
3093 * @cq_period: max period of time in usec before triggering an event
3096 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3099 * ib_destroy_cq - Destroys the specified CQ.
3100 * @cq: The CQ to destroy.
3102 int ib_destroy_cq(struct ib_cq *cq);
3105 * ib_poll_cq - poll a CQ for completion(s)
3106 * @cq:the CQ being polled
3107 * @num_entries:maximum number of completions to return
3108 * @wc:array of at least @num_entries &struct ib_wc where completions
3111 * Poll a CQ for (possibly multiple) completions. If the return value
3112 * is < 0, an error occurred. If the return value is >= 0, it is the
3113 * number of completions returned. If the return value is
3114 * non-negative and < num_entries, then the CQ was emptied.
3116 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3119 return cq->device->poll_cq(cq, num_entries, wc);
3123 * ib_peek_cq - Returns the number of unreaped completions currently
3124 * on the specified CQ.
3125 * @cq: The CQ to peek.
3126 * @wc_cnt: A minimum number of unreaped completions to check for.
3128 * If the number of unreaped completions is greater than or equal to wc_cnt,
3129 * this function returns wc_cnt, otherwise, it returns the actual number of
3130 * unreaped completions.
3132 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3135 * ib_req_notify_cq - Request completion notification on a CQ.
3136 * @cq: The CQ to generate an event for.
3138 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3139 * to request an event on the next solicited event or next work
3140 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3141 * may also be |ed in to request a hint about missed events, as
3145 * < 0 means an error occurred while requesting notification
3146 * == 0 means notification was requested successfully, and if
3147 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3148 * were missed and it is safe to wait for another event. In
3149 * this case is it guaranteed that any work completions added
3150 * to the CQ since the last CQ poll will trigger a completion
3151 * notification event.
3152 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3153 * in. It means that the consumer must poll the CQ again to
3154 * make sure it is empty to avoid missing an event because of a
3155 * race between requesting notification and an entry being
3156 * added to the CQ. This return value means it is possible
3157 * (but not guaranteed) that a work completion has been added
3158 * to the CQ since the last poll without triggering a
3159 * completion notification event.
3161 static inline int ib_req_notify_cq(struct ib_cq *cq,
3162 enum ib_cq_notify_flags flags)
3164 return cq->device->req_notify_cq(cq, flags);
3168 * ib_req_ncomp_notif - Request completion notification when there are
3169 * at least the specified number of unreaped completions on the CQ.
3170 * @cq: The CQ to generate an event for.
3171 * @wc_cnt: The number of unreaped completions that should be on the
3172 * CQ before an event is generated.
3174 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3176 return cq->device->req_ncomp_notif ?
3177 cq->device->req_ncomp_notif(cq, wc_cnt) :
3182 * ib_dma_mapping_error - check a DMA addr for error
3183 * @dev: The device for which the dma_addr was created
3184 * @dma_addr: The DMA address to check
3186 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3188 return dma_mapping_error(dev->dma_device, dma_addr);
3192 * ib_dma_map_single - Map a kernel virtual address to DMA address
3193 * @dev: The device for which the dma_addr is to be created
3194 * @cpu_addr: The kernel virtual address
3195 * @size: The size of the region in bytes
3196 * @direction: The direction of the DMA
3198 static inline u64 ib_dma_map_single(struct ib_device *dev,
3199 void *cpu_addr, size_t size,
3200 enum dma_data_direction direction)
3202 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3206 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3207 * @dev: The device for which the DMA address was created
3208 * @addr: The DMA address
3209 * @size: The size of the region in bytes
3210 * @direction: The direction of the DMA
3212 static inline void ib_dma_unmap_single(struct ib_device *dev,
3213 u64 addr, size_t size,
3214 enum dma_data_direction direction)
3216 dma_unmap_single(dev->dma_device, addr, size, direction);
3220 * ib_dma_map_page - Map a physical page to DMA address
3221 * @dev: The device for which the dma_addr is to be created
3222 * @page: The page to be mapped
3223 * @offset: The offset within the page
3224 * @size: The size of the region in bytes
3225 * @direction: The direction of the DMA
3227 static inline u64 ib_dma_map_page(struct ib_device *dev,
3229 unsigned long offset,
3231 enum dma_data_direction direction)
3233 return dma_map_page(dev->dma_device, page, offset, size, direction);
3237 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3238 * @dev: The device for which the DMA address was created
3239 * @addr: The DMA address
3240 * @size: The size of the region in bytes
3241 * @direction: The direction of the DMA
3243 static inline void ib_dma_unmap_page(struct ib_device *dev,
3244 u64 addr, size_t size,
3245 enum dma_data_direction direction)
3247 dma_unmap_page(dev->dma_device, addr, size, direction);
3251 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3252 * @dev: The device for which the DMA addresses are to be created
3253 * @sg: The array of scatter/gather entries
3254 * @nents: The number of scatter/gather entries
3255 * @direction: The direction of the DMA
3257 static inline int ib_dma_map_sg(struct ib_device *dev,
3258 struct scatterlist *sg, int nents,
3259 enum dma_data_direction direction)
3261 return dma_map_sg(dev->dma_device, sg, nents, direction);
3265 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3266 * @dev: The device for which the DMA addresses were created
3267 * @sg: The array of scatter/gather entries
3268 * @nents: The number of scatter/gather entries
3269 * @direction: The direction of the DMA
3271 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3272 struct scatterlist *sg, int nents,
3273 enum dma_data_direction direction)
3275 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3278 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3279 struct scatterlist *sg, int nents,
3280 enum dma_data_direction direction,
3281 unsigned long dma_attrs)
3283 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3287 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3288 struct scatterlist *sg, int nents,
3289 enum dma_data_direction direction,
3290 unsigned long dma_attrs)
3292 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3295 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3296 * @dev: The device for which the DMA addresses were created
3297 * @sg: The scatter/gather entry
3299 * Note: this function is obsolete. To do: change all occurrences of
3300 * ib_sg_dma_address() into sg_dma_address().
3302 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3303 struct scatterlist *sg)
3305 return sg_dma_address(sg);
3309 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3310 * @dev: The device for which the DMA addresses were created
3311 * @sg: The scatter/gather entry
3313 * Note: this function is obsolete. To do: change all occurrences of
3314 * ib_sg_dma_len() into sg_dma_len().
3316 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3317 struct scatterlist *sg)
3319 return sg_dma_len(sg);
3323 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3324 * @dev: The device for which the DMA address was created
3325 * @addr: The DMA address
3326 * @size: The size of the region in bytes
3327 * @dir: The direction of the DMA
3329 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3332 enum dma_data_direction dir)
3334 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3338 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3339 * @dev: The device for which the DMA address was created
3340 * @addr: The DMA address
3341 * @size: The size of the region in bytes
3342 * @dir: The direction of the DMA
3344 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3347 enum dma_data_direction dir)
3349 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3353 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3354 * @dev: The device for which the DMA address is requested
3355 * @size: The size of the region to allocate in bytes
3356 * @dma_handle: A pointer for returning the DMA address of the region
3357 * @flag: memory allocator flags
3359 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3361 dma_addr_t *dma_handle,
3364 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3368 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3369 * @dev: The device for which the DMA addresses were allocated
3370 * @size: The size of the region
3371 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3372 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3374 static inline void ib_dma_free_coherent(struct ib_device *dev,
3375 size_t size, void *cpu_addr,
3376 dma_addr_t dma_handle)
3378 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3382 * ib_dereg_mr - Deregisters a memory region and removes it from the
3383 * HCA translation table.
3384 * @mr: The memory region to deregister.
3386 * This function can fail, if the memory region has memory windows bound to it.
3388 int ib_dereg_mr(struct ib_mr *mr);
3390 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3391 enum ib_mr_type mr_type,
3395 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3397 * @mr - struct ib_mr pointer to be updated.
3398 * @newkey - new key to be used.
3400 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3402 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3403 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3407 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3408 * for calculating a new rkey for type 2 memory windows.
3409 * @rkey - the rkey to increment.
3411 static inline u32 ib_inc_rkey(u32 rkey)
3413 const u32 mask = 0x000000ff;
3414 return ((rkey + 1) & mask) | (rkey & ~mask);
3418 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3419 * @pd: The protection domain associated with the unmapped region.
3420 * @mr_access_flags: Specifies the memory access rights.
3421 * @fmr_attr: Attributes of the unmapped region.
3423 * A fast memory region must be mapped before it can be used as part of
3426 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3427 int mr_access_flags,
3428 struct ib_fmr_attr *fmr_attr);
3431 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3432 * @fmr: The fast memory region to associate with the pages.
3433 * @page_list: An array of physical pages to map to the fast memory region.
3434 * @list_len: The number of pages in page_list.
3435 * @iova: The I/O virtual address to use with the mapped region.
3437 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3438 u64 *page_list, int list_len,
3441 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3445 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3446 * @fmr_list: A linked list of fast memory regions to unmap.
3448 int ib_unmap_fmr(struct list_head *fmr_list);
3451 * ib_dealloc_fmr - Deallocates a fast memory region.
3452 * @fmr: The fast memory region to deallocate.
3454 int ib_dealloc_fmr(struct ib_fmr *fmr);
3457 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3458 * @qp: QP to attach to the multicast group. The QP must be type
3460 * @gid: Multicast group GID.
3461 * @lid: Multicast group LID in host byte order.
3463 * In order to send and receive multicast packets, subnet
3464 * administration must have created the multicast group and configured
3465 * the fabric appropriately. The port associated with the specified
3466 * QP must also be a member of the multicast group.
3468 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3471 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3472 * @qp: QP to detach from the multicast group.
3473 * @gid: Multicast group GID.
3474 * @lid: Multicast group LID in host byte order.
3476 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3479 * ib_alloc_xrcd - Allocates an XRC domain.
3480 * @device: The device on which to allocate the XRC domain.
3482 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3485 * ib_dealloc_xrcd - Deallocates an XRC domain.
3486 * @xrcd: The XRC domain to deallocate.
3488 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3490 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3491 struct ib_flow_attr *flow_attr, int domain);
3492 int ib_destroy_flow(struct ib_flow *flow_id);
3494 static inline int ib_check_mr_access(int flags)
3497 * Local write permission is required if remote write or
3498 * remote atomic permission is also requested.
3500 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3501 !(flags & IB_ACCESS_LOCAL_WRITE))
3508 * ib_check_mr_status: lightweight check of MR status.
3509 * This routine may provide status checks on a selected
3510 * ib_mr. first use is for signature status check.
3512 * @mr: A memory region.
3513 * @check_mask: Bitmask of which checks to perform from
3514 * ib_mr_status_check enumeration.
3515 * @mr_status: The container of relevant status checks.
3516 * failed checks will be indicated in the status bitmask
3517 * and the relevant info shall be in the error item.
3519 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3520 struct ib_mr_status *mr_status);
3522 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3523 u16 pkey, const union ib_gid *gid,
3524 const struct sockaddr *addr);
3525 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3526 struct ib_wq_init_attr *init_attr);
3527 int ib_destroy_wq(struct ib_wq *wq);
3528 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3530 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3531 struct ib_rwq_ind_table_init_attr*
3532 wq_ind_table_init_attr);
3533 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3535 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3536 unsigned int *sg_offset, unsigned int page_size);
3539 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3540 unsigned int *sg_offset, unsigned int page_size)
3544 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3550 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3551 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3553 void ib_drain_rq(struct ib_qp *qp);
3554 void ib_drain_sq(struct ib_qp *qp);
3555 void ib_drain_qp(struct ib_qp *qp);
3557 int ib_resolve_eth_dmac(struct ib_device *device,
3558 struct rdma_ah_attr *ah_attr);
3560 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3562 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3563 return attr->roce.dmac;
3567 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3569 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3570 attr->ib.dlid = (u16)dlid;
3571 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3572 attr->opa.dlid = dlid;
3575 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3577 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3578 return attr->ib.dlid;
3579 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3580 return attr->opa.dlid;
3584 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3589 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3594 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3597 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3598 attr->ib.src_path_bits = src_path_bits;
3599 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3600 attr->opa.src_path_bits = src_path_bits;
3603 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3605 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3606 return attr->ib.src_path_bits;
3607 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3608 return attr->opa.src_path_bits;
3612 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3614 attr->port_num = port_num;
3617 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3619 return attr->port_num;
3622 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3625 attr->static_rate = static_rate;
3628 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3630 return attr->static_rate;
3633 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3634 enum ib_ah_flags flag)
3636 attr->ah_flags = flag;
3639 static inline enum ib_ah_flags
3640 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3642 return attr->ah_flags;
3645 static inline const struct ib_global_route
3646 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3651 /*To retrieve and modify the grh */
3652 static inline struct ib_global_route
3653 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3658 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3660 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3662 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3665 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3668 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3670 grh->dgid.global.subnet_prefix = prefix;
3673 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3676 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3678 grh->dgid.global.interface_id = if_id;
3681 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3682 union ib_gid *dgid, u32 flow_label,
3683 u8 sgid_index, u8 hop_limit,
3686 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3688 attr->ah_flags = IB_AH_GRH;
3691 grh->flow_label = flow_label;
3692 grh->sgid_index = sgid_index;
3693 grh->hop_limit = hop_limit;
3694 grh->traffic_class = traffic_class;
3698 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3701 if ((rdma_protocol_roce(dev, port_num)) ||
3702 (rdma_protocol_iwarp(dev, port_num)))
3703 return RDMA_AH_ATTR_TYPE_ROCE;
3704 else if ((rdma_protocol_ib(dev, port_num)) &&
3705 (rdma_cap_opa_ah(dev, port_num)))
3706 return RDMA_AH_ATTR_TYPE_OPA;
3708 return RDMA_AH_ATTR_TYPE_IB;
3710 #endif /* IB_VERBS_H */