2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
68 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
70 extern struct workqueue_struct *ib_wq;
71 extern struct workqueue_struct *ib_comp_wq;
81 extern union ib_gid zgid;
84 /* If link layer is Ethernet, this is RoCE V1 */
87 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
91 #define ROCE_V2_UDP_DPORT 4791
93 enum ib_gid_type gid_type;
94 struct net_device *ndev;
98 /* IB values map to NodeInfo:NodeType. */
108 /* set the local administered indication */
109 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
112 enum rdma_transport_type {
114 RDMA_TRANSPORT_IWARP,
115 RDMA_TRANSPORT_USNIC,
116 RDMA_TRANSPORT_USNIC_UDP
119 enum rdma_protocol_type {
123 RDMA_PROTOCOL_USNIC_UDP
126 __attribute_const__ enum rdma_transport_type
127 rdma_node_get_transport(enum rdma_node_type node_type);
129 enum rdma_network_type {
131 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
136 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
138 if (network_type == RDMA_NETWORK_IPV4 ||
139 network_type == RDMA_NETWORK_IPV6)
140 return IB_GID_TYPE_ROCE_UDP_ENCAP;
142 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
143 return IB_GID_TYPE_IB;
146 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
149 if (gid_type == IB_GID_TYPE_IB)
150 return RDMA_NETWORK_IB;
152 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
153 return RDMA_NETWORK_IPV4;
155 return RDMA_NETWORK_IPV6;
158 enum rdma_link_layer {
159 IB_LINK_LAYER_UNSPECIFIED,
160 IB_LINK_LAYER_INFINIBAND,
161 IB_LINK_LAYER_ETHERNET,
164 enum ib_device_cap_flags {
165 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
166 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
167 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
168 IB_DEVICE_RAW_MULTI = (1 << 3),
169 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
170 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
171 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
172 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
173 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
174 /* Not in use, former INIT_TYPE = (1 << 9),*/
175 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
176 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
177 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
178 IB_DEVICE_SRQ_RESIZE = (1 << 13),
179 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
182 * This device supports a per-device lkey or stag that can be
183 * used without performing a memory registration for the local
184 * memory. Note that ULPs should never check this flag, but
185 * instead of use the local_dma_lkey flag in the ib_pd structure,
186 * which will always contain a usable lkey.
188 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
189 /* Reserved, old SEND_W_INV = (1 << 16),*/
190 IB_DEVICE_MEM_WINDOW = (1 << 17),
192 * Devices should set IB_DEVICE_UD_IP_SUM if they support
193 * insertion of UDP and TCP checksum on outgoing UD IPoIB
194 * messages and can verify the validity of checksum for
195 * incoming messages. Setting this flag implies that the
196 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
198 IB_DEVICE_UD_IP_CSUM = (1 << 18),
199 IB_DEVICE_UD_TSO = (1 << 19),
200 IB_DEVICE_XRC = (1 << 20),
203 * This device supports the IB "base memory management extension",
204 * which includes support for fast registrations (IB_WR_REG_MR,
205 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
206 * also be set by any iWarp device which must support FRs to comply
207 * to the iWarp verbs spec. iWarp devices also support the
208 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
211 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
212 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
213 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
214 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
215 IB_DEVICE_RC_IP_CSUM = (1 << 25),
216 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
217 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
219 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 * support execution of WQEs that involve synchronization
221 * of I/O operations with single completion queue managed
224 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
225 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
226 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
227 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
228 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
229 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
230 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
231 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
232 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
233 /* The device supports padding incoming writes to cacheline. */
234 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
237 enum ib_signature_prot_cap {
238 IB_PROT_T10DIF_TYPE_1 = 1,
239 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
240 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
243 enum ib_signature_guard_cap {
244 IB_GUARD_T10DIF_CRC = 1,
245 IB_GUARD_T10DIF_CSUM = 1 << 1,
254 enum ib_odp_general_cap_bits {
255 IB_ODP_SUPPORT = 1 << 0,
256 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
259 enum ib_odp_transport_cap_bits {
260 IB_ODP_SUPPORT_SEND = 1 << 0,
261 IB_ODP_SUPPORT_RECV = 1 << 1,
262 IB_ODP_SUPPORT_WRITE = 1 << 2,
263 IB_ODP_SUPPORT_READ = 1 << 3,
264 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
268 uint64_t general_caps;
270 uint32_t rc_odp_caps;
271 uint32_t uc_odp_caps;
272 uint32_t ud_odp_caps;
273 } per_transport_caps;
277 /* Corresponding bit will be set if qp type from
278 * 'enum ib_qp_type' is supported, e.g.
279 * supported_qpts |= 1 << IB_QPT_UD
282 u32 max_rwq_indirection_tables;
283 u32 max_rwq_indirection_table_size;
286 enum ib_tm_cap_flags {
287 /* Support tag matching on RC transport */
288 IB_TM_CAP_RC = 1 << 0,
292 /* Max size of RNDV header */
293 u32 max_rndv_hdr_size;
294 /* Max number of entries in tag matching list */
296 /* From enum ib_tm_cap_flags */
298 /* Max number of outstanding list operations */
300 /* Max number of SGE in tag matching entry */
304 struct ib_cq_init_attr {
310 enum ib_cq_attr_mask {
311 IB_CQ_MODERATE = 1 << 0,
315 u16 max_cq_moderation_count;
316 u16 max_cq_moderation_period;
319 struct ib_device_attr {
321 __be64 sys_image_guid;
329 u64 device_cap_flags;
339 int max_qp_init_rd_atom;
340 int max_ee_init_rd_atom;
341 enum ib_atomic_cap atomic_cap;
342 enum ib_atomic_cap masked_atomic_cap;
349 int max_mcast_qp_attach;
350 int max_total_mcast_qp_attach;
357 unsigned int max_fast_reg_page_list_len;
359 u8 local_ca_ack_delay;
362 struct ib_odp_caps odp_caps;
363 uint64_t timestamp_mask;
364 uint64_t hca_core_clock; /* in KHZ */
365 struct ib_rss_caps rss_caps;
367 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
368 struct ib_tm_caps tm_caps;
369 struct ib_cq_caps cq_caps;
380 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
383 case IB_MTU_256: return 256;
384 case IB_MTU_512: return 512;
385 case IB_MTU_1024: return 1024;
386 case IB_MTU_2048: return 2048;
387 case IB_MTU_4096: return 4096;
392 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
396 else if (mtu >= 2048)
398 else if (mtu >= 1024)
412 IB_PORT_ACTIVE_DEFER = 5
415 enum ib_port_cap_flags {
417 IB_PORT_NOTICE_SUP = 1 << 2,
418 IB_PORT_TRAP_SUP = 1 << 3,
419 IB_PORT_OPT_IPD_SUP = 1 << 4,
420 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
421 IB_PORT_SL_MAP_SUP = 1 << 6,
422 IB_PORT_MKEY_NVRAM = 1 << 7,
423 IB_PORT_PKEY_NVRAM = 1 << 8,
424 IB_PORT_LED_INFO_SUP = 1 << 9,
425 IB_PORT_SM_DISABLED = 1 << 10,
426 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
427 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
428 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
429 IB_PORT_CM_SUP = 1 << 16,
430 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
431 IB_PORT_REINIT_SUP = 1 << 18,
432 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
433 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
434 IB_PORT_DR_NOTICE_SUP = 1 << 21,
435 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
436 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
437 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
438 IB_PORT_CLIENT_REG_SUP = 1 << 25,
439 IB_PORT_IP_BASED_GIDS = 1 << 26,
449 static inline int ib_width_enum_to_int(enum ib_port_width width)
452 case IB_WIDTH_1X: return 1;
453 case IB_WIDTH_4X: return 4;
454 case IB_WIDTH_8X: return 8;
455 case IB_WIDTH_12X: return 12;
471 * struct rdma_hw_stats
472 * @timestamp - Used by the core code to track when the last update was
473 * @lifespan - Used by the core code to determine how old the counters
474 * should be before being updated again. Stored in jiffies, defaults
475 * to 10 milliseconds, drivers can override the default be specifying
476 * their own value during their allocation routine.
477 * @name - Array of pointers to static names used for the counters in
479 * @num_counters - How many hardware counters there are. If name is
480 * shorter than this number, a kernel oops will result. Driver authors
481 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
482 * in their code to prevent this.
483 * @value - Array of u64 counters that are accessed by the sysfs code and
484 * filled in by the drivers get_stats routine
486 struct rdma_hw_stats {
487 unsigned long timestamp;
488 unsigned long lifespan;
489 const char * const *names;
494 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
496 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
498 * @names - Array of static const char *
499 * @num_counters - How many elements in array
500 * @lifespan - How many milliseconds between updates
502 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
503 const char * const *names, int num_counters,
504 unsigned long lifespan)
506 struct rdma_hw_stats *stats;
508 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
512 stats->names = names;
513 stats->num_counters = num_counters;
514 stats->lifespan = msecs_to_jiffies(lifespan);
520 /* Define bits for the various functionality this port needs to be supported by
523 /* Management 0x00000FFF */
524 #define RDMA_CORE_CAP_IB_MAD 0x00000001
525 #define RDMA_CORE_CAP_IB_SMI 0x00000002
526 #define RDMA_CORE_CAP_IB_CM 0x00000004
527 #define RDMA_CORE_CAP_IW_CM 0x00000008
528 #define RDMA_CORE_CAP_IB_SA 0x00000010
529 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
531 /* Address format 0x000FF000 */
532 #define RDMA_CORE_CAP_AF_IB 0x00001000
533 #define RDMA_CORE_CAP_ETH_AH 0x00002000
534 #define RDMA_CORE_CAP_OPA_AH 0x00004000
536 /* Protocol 0xFFF00000 */
537 #define RDMA_CORE_CAP_PROT_IB 0x00100000
538 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
539 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
541 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
542 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
544 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
545 | RDMA_CORE_CAP_IB_MAD \
546 | RDMA_CORE_CAP_IB_SMI \
547 | RDMA_CORE_CAP_IB_CM \
548 | RDMA_CORE_CAP_IB_SA \
549 | RDMA_CORE_CAP_AF_IB)
550 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
551 | RDMA_CORE_CAP_IB_MAD \
552 | RDMA_CORE_CAP_IB_CM \
553 | RDMA_CORE_CAP_AF_IB \
554 | RDMA_CORE_CAP_ETH_AH)
555 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
556 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
557 | RDMA_CORE_CAP_IB_MAD \
558 | RDMA_CORE_CAP_IB_CM \
559 | RDMA_CORE_CAP_AF_IB \
560 | RDMA_CORE_CAP_ETH_AH)
561 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
562 | RDMA_CORE_CAP_IW_CM)
563 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
564 | RDMA_CORE_CAP_OPA_MAD)
566 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
568 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
570 struct ib_port_attr {
572 enum ib_port_state state;
574 enum ib_mtu active_mtu;
594 enum ib_device_modify_flags {
595 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
596 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
599 #define IB_DEVICE_NODE_DESC_MAX 64
601 struct ib_device_modify {
603 char node_desc[IB_DEVICE_NODE_DESC_MAX];
606 enum ib_port_modify_flags {
607 IB_PORT_SHUTDOWN = 1,
608 IB_PORT_INIT_TYPE = (1<<2),
609 IB_PORT_RESET_QKEY_CNTR = (1<<3),
610 IB_PORT_OPA_MASK_CHG = (1<<4)
613 struct ib_port_modify {
614 u32 set_port_cap_mask;
615 u32 clr_port_cap_mask;
623 IB_EVENT_QP_ACCESS_ERR,
627 IB_EVENT_PATH_MIG_ERR,
628 IB_EVENT_DEVICE_FATAL,
629 IB_EVENT_PORT_ACTIVE,
632 IB_EVENT_PKEY_CHANGE,
635 IB_EVENT_SRQ_LIMIT_REACHED,
636 IB_EVENT_QP_LAST_WQE_REACHED,
637 IB_EVENT_CLIENT_REREGISTER,
642 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
645 struct ib_device *device;
653 enum ib_event_type event;
656 struct ib_event_handler {
657 struct ib_device *device;
658 void (*handler)(struct ib_event_handler *, struct ib_event *);
659 struct list_head list;
662 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
664 (_ptr)->device = _device; \
665 (_ptr)->handler = _handler; \
666 INIT_LIST_HEAD(&(_ptr)->list); \
669 struct ib_global_route {
678 __be32 version_tclass_flow;
686 union rdma_network_hdr {
689 /* The IB spec states that if it's IPv4, the header
690 * is located in the last 20 bytes of the header.
693 struct iphdr roce4grh;
697 #define IB_QPN_MASK 0xFFFFFF
700 IB_MULTICAST_QPN = 0xffffff
703 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
704 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
711 IB_RATE_PORT_CURRENT = 0,
712 IB_RATE_2_5_GBPS = 2,
720 IB_RATE_120_GBPS = 10,
721 IB_RATE_14_GBPS = 11,
722 IB_RATE_56_GBPS = 12,
723 IB_RATE_112_GBPS = 13,
724 IB_RATE_168_GBPS = 14,
725 IB_RATE_25_GBPS = 15,
726 IB_RATE_100_GBPS = 16,
727 IB_RATE_200_GBPS = 17,
728 IB_RATE_300_GBPS = 18
732 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
733 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
734 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
735 * @rate: rate to convert.
737 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
740 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
741 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
742 * @rate: rate to convert.
744 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
748 * enum ib_mr_type - memory region type
749 * @IB_MR_TYPE_MEM_REG: memory region that is used for
750 * normal registration
751 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
752 * signature operations (data-integrity
754 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
755 * register any arbitrary sg lists (without
756 * the normal mr constraints - see
761 IB_MR_TYPE_SIGNATURE,
767 * IB_SIG_TYPE_NONE: Unprotected.
768 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
770 enum ib_signature_type {
776 * Signature T10-DIF block-guard types
777 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
778 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
780 enum ib_t10_dif_bg_type {
786 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
788 * @bg_type: T10-DIF block guard type (CRC|CSUM)
789 * @pi_interval: protection information interval.
790 * @bg: seed of guard computation.
791 * @app_tag: application tag of guard block
792 * @ref_tag: initial guard block reference tag.
793 * @ref_remap: Indicate wethear the reftag increments each block
794 * @app_escape: Indicate to skip block check if apptag=0xffff
795 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
796 * @apptag_check_mask: check bitmask of application tag.
798 struct ib_t10_dif_domain {
799 enum ib_t10_dif_bg_type bg_type;
807 u16 apptag_check_mask;
811 * struct ib_sig_domain - Parameters for signature domain
812 * @sig_type: specific signauture type
813 * @sig: union of all signature domain attributes that may
814 * be used to set domain layout.
816 struct ib_sig_domain {
817 enum ib_signature_type sig_type;
819 struct ib_t10_dif_domain dif;
824 * struct ib_sig_attrs - Parameters for signature handover operation
825 * @check_mask: bitmask for signature byte check (8 bytes)
826 * @mem: memory domain layout desciptor.
827 * @wire: wire domain layout desciptor.
829 struct ib_sig_attrs {
831 struct ib_sig_domain mem;
832 struct ib_sig_domain wire;
835 enum ib_sig_err_type {
842 * struct ib_sig_err - signature error descriptor
845 enum ib_sig_err_type err_type;
852 enum ib_mr_status_check {
853 IB_MR_CHECK_SIG_STATUS = 1,
857 * struct ib_mr_status - Memory region status container
859 * @fail_status: Bitmask of MR checks status. For each
860 * failed check a corresponding status bit is set.
861 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
864 struct ib_mr_status {
866 struct ib_sig_err sig_err;
870 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
872 * @mult: multiple to convert.
874 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
876 enum rdma_ah_attr_type {
877 RDMA_AH_ATTR_TYPE_UNDEFINED,
878 RDMA_AH_ATTR_TYPE_IB,
879 RDMA_AH_ATTR_TYPE_ROCE,
880 RDMA_AH_ATTR_TYPE_OPA,
888 struct roce_ah_attr {
898 struct rdma_ah_attr {
899 struct ib_global_route grh;
904 enum rdma_ah_attr_type type;
906 struct ib_ah_attr ib;
907 struct roce_ah_attr roce;
908 struct opa_ah_attr opa;
916 IB_WC_LOC_EEC_OP_ERR,
921 IB_WC_LOC_ACCESS_ERR,
922 IB_WC_REM_INV_REQ_ERR,
923 IB_WC_REM_ACCESS_ERR,
926 IB_WC_RNR_RETRY_EXC_ERR,
927 IB_WC_LOC_RDD_VIOL_ERR,
928 IB_WC_REM_INV_RD_REQ_ERR,
931 IB_WC_INV_EEC_STATE_ERR,
933 IB_WC_RESP_TIMEOUT_ERR,
937 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
948 IB_WC_MASKED_COMP_SWAP,
949 IB_WC_MASKED_FETCH_ADD,
951 * Set value of IB_WC_RECV so consumers can test if a completion is a
952 * receive by testing (opcode & IB_WC_RECV).
955 IB_WC_RECV_RDMA_WITH_IMM
960 IB_WC_WITH_IMM = (1<<1),
961 IB_WC_WITH_INVALIDATE = (1<<2),
962 IB_WC_IP_CSUM_OK = (1<<3),
963 IB_WC_WITH_SMAC = (1<<4),
964 IB_WC_WITH_VLAN = (1<<5),
965 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
971 struct ib_cqe *wr_cqe;
973 enum ib_wc_status status;
974 enum ib_wc_opcode opcode;
988 u8 port_num; /* valid only for DR SMPs on switches */
994 enum ib_cq_notify_flags {
995 IB_CQ_SOLICITED = 1 << 0,
996 IB_CQ_NEXT_COMP = 1 << 1,
997 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
998 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1007 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1009 return srq_type == IB_SRQT_XRC ||
1010 srq_type == IB_SRQT_TM;
1013 enum ib_srq_attr_mask {
1014 IB_SRQ_MAX_WR = 1 << 0,
1015 IB_SRQ_LIMIT = 1 << 1,
1018 struct ib_srq_attr {
1024 struct ib_srq_init_attr {
1025 void (*event_handler)(struct ib_event *, void *);
1027 struct ib_srq_attr attr;
1028 enum ib_srq_type srq_type;
1034 struct ib_xrcd *xrcd;
1049 u32 max_inline_data;
1052 * Maximum number of rdma_rw_ctx structures in flight at a time.
1053 * ib_create_qp() will calculate the right amount of neededed WRs
1054 * and MRs based on this.
1066 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1067 * here (and in that order) since the MAD layer uses them as
1068 * indices into a 2-entry table.
1077 IB_QPT_RAW_ETHERTYPE,
1078 IB_QPT_RAW_PACKET = 8,
1082 IB_QPT_DRIVER = 0xFF,
1083 /* Reserve a range for qp types internal to the low level driver.
1084 * These qp types will not be visible at the IB core layer, so the
1085 * IB_QPT_MAX usages should not be affected in the core layer
1087 IB_QPT_RESERVED1 = 0x1000,
1099 enum ib_qp_create_flags {
1100 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1101 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1102 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1103 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1104 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1105 IB_QP_CREATE_NETIF_QP = 1 << 5,
1106 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1107 /* FREE = 1 << 7, */
1108 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1109 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1110 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1111 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1112 /* reserve bits 26-31 for low level drivers' internal use */
1113 IB_QP_CREATE_RESERVED_START = 1 << 26,
1114 IB_QP_CREATE_RESERVED_END = 1 << 31,
1118 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1119 * callback to destroy the passed in QP.
1122 struct ib_qp_init_attr {
1123 void (*event_handler)(struct ib_event *, void *);
1125 struct ib_cq *send_cq;
1126 struct ib_cq *recv_cq;
1128 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1129 struct ib_qp_cap cap;
1130 enum ib_sig_type sq_sig_type;
1131 enum ib_qp_type qp_type;
1132 enum ib_qp_create_flags create_flags;
1135 * Only needed for special QP types, or when using the RW API.
1138 struct ib_rwq_ind_table *rwq_ind_tbl;
1142 struct ib_qp_open_attr {
1143 void (*event_handler)(struct ib_event *, void *);
1146 enum ib_qp_type qp_type;
1149 enum ib_rnr_timeout {
1150 IB_RNR_TIMER_655_36 = 0,
1151 IB_RNR_TIMER_000_01 = 1,
1152 IB_RNR_TIMER_000_02 = 2,
1153 IB_RNR_TIMER_000_03 = 3,
1154 IB_RNR_TIMER_000_04 = 4,
1155 IB_RNR_TIMER_000_06 = 5,
1156 IB_RNR_TIMER_000_08 = 6,
1157 IB_RNR_TIMER_000_12 = 7,
1158 IB_RNR_TIMER_000_16 = 8,
1159 IB_RNR_TIMER_000_24 = 9,
1160 IB_RNR_TIMER_000_32 = 10,
1161 IB_RNR_TIMER_000_48 = 11,
1162 IB_RNR_TIMER_000_64 = 12,
1163 IB_RNR_TIMER_000_96 = 13,
1164 IB_RNR_TIMER_001_28 = 14,
1165 IB_RNR_TIMER_001_92 = 15,
1166 IB_RNR_TIMER_002_56 = 16,
1167 IB_RNR_TIMER_003_84 = 17,
1168 IB_RNR_TIMER_005_12 = 18,
1169 IB_RNR_TIMER_007_68 = 19,
1170 IB_RNR_TIMER_010_24 = 20,
1171 IB_RNR_TIMER_015_36 = 21,
1172 IB_RNR_TIMER_020_48 = 22,
1173 IB_RNR_TIMER_030_72 = 23,
1174 IB_RNR_TIMER_040_96 = 24,
1175 IB_RNR_TIMER_061_44 = 25,
1176 IB_RNR_TIMER_081_92 = 26,
1177 IB_RNR_TIMER_122_88 = 27,
1178 IB_RNR_TIMER_163_84 = 28,
1179 IB_RNR_TIMER_245_76 = 29,
1180 IB_RNR_TIMER_327_68 = 30,
1181 IB_RNR_TIMER_491_52 = 31
1184 enum ib_qp_attr_mask {
1186 IB_QP_CUR_STATE = (1<<1),
1187 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1188 IB_QP_ACCESS_FLAGS = (1<<3),
1189 IB_QP_PKEY_INDEX = (1<<4),
1190 IB_QP_PORT = (1<<5),
1191 IB_QP_QKEY = (1<<6),
1193 IB_QP_PATH_MTU = (1<<8),
1194 IB_QP_TIMEOUT = (1<<9),
1195 IB_QP_RETRY_CNT = (1<<10),
1196 IB_QP_RNR_RETRY = (1<<11),
1197 IB_QP_RQ_PSN = (1<<12),
1198 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1199 IB_QP_ALT_PATH = (1<<14),
1200 IB_QP_MIN_RNR_TIMER = (1<<15),
1201 IB_QP_SQ_PSN = (1<<16),
1202 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1203 IB_QP_PATH_MIG_STATE = (1<<18),
1204 IB_QP_CAP = (1<<19),
1205 IB_QP_DEST_QPN = (1<<20),
1206 IB_QP_RESERVED1 = (1<<21),
1207 IB_QP_RESERVED2 = (1<<22),
1208 IB_QP_RESERVED3 = (1<<23),
1209 IB_QP_RESERVED4 = (1<<24),
1210 IB_QP_RATE_LIMIT = (1<<25),
1235 enum ib_qp_state qp_state;
1236 enum ib_qp_state cur_qp_state;
1237 enum ib_mtu path_mtu;
1238 enum ib_mig_state path_mig_state;
1243 int qp_access_flags;
1244 struct ib_qp_cap cap;
1245 struct rdma_ah_attr ah_attr;
1246 struct rdma_ah_attr alt_ah_attr;
1249 u8 en_sqd_async_notify;
1252 u8 max_dest_rd_atomic;
1265 IB_WR_RDMA_WRITE_WITH_IMM,
1267 IB_WR_SEND_WITH_IMM,
1269 IB_WR_ATOMIC_CMP_AND_SWP,
1270 IB_WR_ATOMIC_FETCH_AND_ADD,
1272 IB_WR_SEND_WITH_INV,
1273 IB_WR_RDMA_READ_WITH_INV,
1276 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1277 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1279 /* reserve values for low level drivers' internal use.
1280 * These values will not be used at all in the ib core layer.
1282 IB_WR_RESERVED1 = 0xf0,
1294 enum ib_send_flags {
1296 IB_SEND_SIGNALED = (1<<1),
1297 IB_SEND_SOLICITED = (1<<2),
1298 IB_SEND_INLINE = (1<<3),
1299 IB_SEND_IP_CSUM = (1<<4),
1301 /* reserve bits 26-31 for low level drivers' internal use */
1302 IB_SEND_RESERVED_START = (1 << 26),
1303 IB_SEND_RESERVED_END = (1 << 31),
1313 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1317 struct ib_send_wr *next;
1320 struct ib_cqe *wr_cqe;
1322 struct ib_sge *sg_list;
1324 enum ib_wr_opcode opcode;
1328 u32 invalidate_rkey;
1333 struct ib_send_wr wr;
1338 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1340 return container_of(wr, struct ib_rdma_wr, wr);
1343 struct ib_atomic_wr {
1344 struct ib_send_wr wr;
1348 u64 compare_add_mask;
1353 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1355 return container_of(wr, struct ib_atomic_wr, wr);
1359 struct ib_send_wr wr;
1366 u16 pkey_index; /* valid for GSI only */
1367 u8 port_num; /* valid for DR SMPs on switch only */
1370 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1372 return container_of(wr, struct ib_ud_wr, wr);
1376 struct ib_send_wr wr;
1382 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1384 return container_of(wr, struct ib_reg_wr, wr);
1387 struct ib_sig_handover_wr {
1388 struct ib_send_wr wr;
1389 struct ib_sig_attrs *sig_attrs;
1390 struct ib_mr *sig_mr;
1392 struct ib_sge *prot;
1395 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1397 return container_of(wr, struct ib_sig_handover_wr, wr);
1401 struct ib_recv_wr *next;
1404 struct ib_cqe *wr_cqe;
1406 struct ib_sge *sg_list;
1410 enum ib_access_flags {
1411 IB_ACCESS_LOCAL_WRITE = 1,
1412 IB_ACCESS_REMOTE_WRITE = (1<<1),
1413 IB_ACCESS_REMOTE_READ = (1<<2),
1414 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1415 IB_ACCESS_MW_BIND = (1<<4),
1416 IB_ZERO_BASED = (1<<5),
1417 IB_ACCESS_ON_DEMAND = (1<<6),
1418 IB_ACCESS_HUGETLB = (1<<7),
1422 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1423 * are hidden here instead of a uapi header!
1425 enum ib_mr_rereg_flags {
1426 IB_MR_REREG_TRANS = 1,
1427 IB_MR_REREG_PD = (1<<1),
1428 IB_MR_REREG_ACCESS = (1<<2),
1429 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1432 struct ib_fmr_attr {
1440 enum rdma_remove_reason {
1441 /* Userspace requested uobject deletion. Call could fail */
1442 RDMA_REMOVE_DESTROY,
1443 /* Context deletion. This call should delete the actual object itself */
1445 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1446 RDMA_REMOVE_DRIVER_REMOVE,
1447 /* Context is being cleaned-up, but commit was just completed */
1448 RDMA_REMOVE_DURING_CLEANUP,
1451 struct ib_rdmacg_object {
1452 #ifdef CONFIG_CGROUP_RDMA
1453 struct rdma_cgroup *cg; /* owner rdma cgroup */
1457 struct ib_ucontext {
1458 struct ib_device *device;
1459 struct ib_uverbs_file *ufile;
1462 /* locking the uobjects_list */
1463 struct mutex uobjects_lock;
1464 struct list_head uobjects;
1465 /* protects cleanup process from other actions */
1466 struct rw_semaphore cleanup_rwsem;
1467 enum rdma_remove_reason cleanup_reason;
1470 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1471 struct rb_root_cached umem_tree;
1473 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1474 * mmu notifiers registration.
1476 struct rw_semaphore umem_rwsem;
1477 void (*invalidate_range)(struct ib_umem *umem,
1478 unsigned long start, unsigned long end);
1480 struct mmu_notifier mn;
1481 atomic_t notifier_count;
1482 /* A list of umems that don't have private mmu notifier counters yet. */
1483 struct list_head no_private_counters;
1487 struct ib_rdmacg_object cg_obj;
1491 u64 user_handle; /* handle given to us by userspace */
1492 struct ib_ucontext *context; /* associated user context */
1493 void *object; /* containing object */
1494 struct list_head list; /* link to context's list */
1495 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1496 int id; /* index into kernel idr */
1498 atomic_t usecnt; /* protects exclusive access */
1499 struct rcu_head rcu; /* kfree_rcu() overhead */
1501 const struct uverbs_obj_type *type;
1504 struct ib_uobject_file {
1505 struct ib_uobject uobj;
1506 /* ufile contains the lock between context release and file close */
1507 struct ib_uverbs_file *ufile;
1511 const void __user *inbuf;
1512 void __user *outbuf;
1520 struct ib_device *device;
1521 struct ib_uobject *uobject;
1522 atomic_t usecnt; /* count all resources */
1524 u32 unsafe_global_rkey;
1527 * Implementation details of the RDMA core, don't use in drivers:
1529 struct ib_mr *__internal_mr;
1530 struct rdma_restrack_entry res;
1534 struct ib_device *device;
1535 atomic_t usecnt; /* count all exposed resources */
1536 struct inode *inode;
1538 struct mutex tgt_qp_mutex;
1539 struct list_head tgt_qp_list;
1543 struct ib_device *device;
1545 struct ib_uobject *uobject;
1546 enum rdma_ah_attr_type type;
1549 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1551 enum ib_poll_context {
1552 IB_POLL_DIRECT, /* caller context, no hw completions */
1553 IB_POLL_SOFTIRQ, /* poll from softirq context */
1554 IB_POLL_WORKQUEUE, /* poll from workqueue */
1558 struct ib_device *device;
1559 struct ib_uobject *uobject;
1560 ib_comp_handler comp_handler;
1561 void (*event_handler)(struct ib_event *, void *);
1564 atomic_t usecnt; /* count number of work queues */
1565 enum ib_poll_context poll_ctx;
1568 struct irq_poll iop;
1569 struct work_struct work;
1572 * Implementation details of the RDMA core, don't use in drivers:
1574 struct rdma_restrack_entry res;
1578 struct ib_device *device;
1580 struct ib_uobject *uobject;
1581 void (*event_handler)(struct ib_event *, void *);
1583 enum ib_srq_type srq_type;
1590 struct ib_xrcd *xrcd;
1597 enum ib_raw_packet_caps {
1598 /* Strip cvlan from incoming packet and report it in the matching work
1599 * completion is supported.
1601 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1602 /* Scatter FCS field of an incoming packet to host memory is supported.
1604 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1605 /* Checksum offloads are supported (for both send and receive). */
1606 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1607 /* When a packet is received for an RQ with no receive WQEs, the
1608 * packet processing is delayed.
1610 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1624 struct ib_device *device;
1625 struct ib_uobject *uobject;
1627 void (*event_handler)(struct ib_event *, void *);
1631 enum ib_wq_state state;
1632 enum ib_wq_type wq_type;
1637 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1638 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1639 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1640 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1643 struct ib_wq_init_attr {
1645 enum ib_wq_type wq_type;
1649 void (*event_handler)(struct ib_event *, void *);
1650 u32 create_flags; /* Use enum ib_wq_flags */
1653 enum ib_wq_attr_mask {
1654 IB_WQ_STATE = 1 << 0,
1655 IB_WQ_CUR_STATE = 1 << 1,
1656 IB_WQ_FLAGS = 1 << 2,
1660 enum ib_wq_state wq_state;
1661 enum ib_wq_state curr_wq_state;
1662 u32 flags; /* Use enum ib_wq_flags */
1663 u32 flags_mask; /* Use enum ib_wq_flags */
1666 struct ib_rwq_ind_table {
1667 struct ib_device *device;
1668 struct ib_uobject *uobject;
1671 u32 log_ind_tbl_size;
1672 struct ib_wq **ind_tbl;
1675 struct ib_rwq_ind_table_init_attr {
1676 u32 log_ind_tbl_size;
1677 /* Each entry is a pointer to Receive Work Queue */
1678 struct ib_wq **ind_tbl;
1681 enum port_pkey_state {
1682 IB_PORT_PKEY_NOT_VALID = 0,
1683 IB_PORT_PKEY_VALID = 1,
1684 IB_PORT_PKEY_LISTED = 2,
1687 struct ib_qp_security;
1689 struct ib_port_pkey {
1690 enum port_pkey_state state;
1693 struct list_head qp_list;
1694 struct list_head to_error_list;
1695 struct ib_qp_security *sec;
1698 struct ib_ports_pkeys {
1699 struct ib_port_pkey main;
1700 struct ib_port_pkey alt;
1703 struct ib_qp_security {
1705 struct ib_device *dev;
1706 /* Hold this mutex when changing port and pkey settings. */
1708 struct ib_ports_pkeys *ports_pkeys;
1709 /* A list of all open shared QP handles. Required to enforce security
1710 * properly for all users of a shared QP.
1712 struct list_head shared_qp_list;
1715 atomic_t error_list_count;
1716 struct completion error_complete;
1717 int error_comps_pending;
1721 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1722 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1725 struct ib_device *device;
1727 struct ib_cq *send_cq;
1728 struct ib_cq *recv_cq;
1731 struct list_head rdma_mrs;
1732 struct list_head sig_mrs;
1734 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1735 struct list_head xrcd_list;
1737 /* count times opened, mcast attaches, flow attaches */
1739 struct list_head open_list;
1740 struct ib_qp *real_qp;
1741 struct ib_uobject *uobject;
1742 void (*event_handler)(struct ib_event *, void *);
1747 enum ib_qp_type qp_type;
1748 struct ib_rwq_ind_table *rwq_ind_tbl;
1749 struct ib_qp_security *qp_sec;
1753 * Implementation details of the RDMA core, don't use in drivers:
1755 struct rdma_restrack_entry res;
1759 struct ib_device *device;
1765 unsigned int page_size;
1768 struct ib_uobject *uobject; /* user */
1769 struct list_head qp_entry; /* FR */
1774 struct ib_device *device;
1776 struct ib_uobject *uobject;
1778 enum ib_mw_type type;
1782 struct ib_device *device;
1784 struct list_head list;
1789 /* Supported steering options */
1790 enum ib_flow_attr_type {
1791 /* steering according to rule specifications */
1792 IB_FLOW_ATTR_NORMAL = 0x0,
1793 /* default unicast and multicast rule -
1794 * receive all Eth traffic which isn't steered to any QP
1796 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1797 /* default multicast rule -
1798 * receive all Eth multicast traffic which isn't steered to any QP
1800 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1801 /* sniffer rule - receive all port traffic */
1802 IB_FLOW_ATTR_SNIFFER = 0x3
1805 /* Supported steering header types */
1806 enum ib_flow_spec_type {
1808 IB_FLOW_SPEC_ETH = 0x20,
1809 IB_FLOW_SPEC_IB = 0x22,
1811 IB_FLOW_SPEC_IPV4 = 0x30,
1812 IB_FLOW_SPEC_IPV6 = 0x31,
1814 IB_FLOW_SPEC_TCP = 0x40,
1815 IB_FLOW_SPEC_UDP = 0x41,
1816 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1817 IB_FLOW_SPEC_INNER = 0x100,
1819 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1820 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1822 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1823 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1825 /* Flow steering rule priority is set according to it's domain.
1826 * Lower domain value means higher priority.
1828 enum ib_flow_domain {
1829 IB_FLOW_DOMAIN_USER,
1830 IB_FLOW_DOMAIN_ETHTOOL,
1833 IB_FLOW_DOMAIN_NUM /* Must be last */
1836 enum ib_flow_flags {
1837 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1838 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1841 struct ib_flow_eth_filter {
1850 struct ib_flow_spec_eth {
1853 struct ib_flow_eth_filter val;
1854 struct ib_flow_eth_filter mask;
1857 struct ib_flow_ib_filter {
1864 struct ib_flow_spec_ib {
1867 struct ib_flow_ib_filter val;
1868 struct ib_flow_ib_filter mask;
1871 /* IPv4 header flags */
1872 enum ib_ipv4_flags {
1873 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1874 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1875 last have this flag set */
1878 struct ib_flow_ipv4_filter {
1889 struct ib_flow_spec_ipv4 {
1892 struct ib_flow_ipv4_filter val;
1893 struct ib_flow_ipv4_filter mask;
1896 struct ib_flow_ipv6_filter {
1907 struct ib_flow_spec_ipv6 {
1910 struct ib_flow_ipv6_filter val;
1911 struct ib_flow_ipv6_filter mask;
1914 struct ib_flow_tcp_udp_filter {
1921 struct ib_flow_spec_tcp_udp {
1924 struct ib_flow_tcp_udp_filter val;
1925 struct ib_flow_tcp_udp_filter mask;
1928 struct ib_flow_tunnel_filter {
1933 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1934 * the tunnel_id from val has the vni value
1936 struct ib_flow_spec_tunnel {
1939 struct ib_flow_tunnel_filter val;
1940 struct ib_flow_tunnel_filter mask;
1943 struct ib_flow_spec_action_tag {
1944 enum ib_flow_spec_type type;
1949 struct ib_flow_spec_action_drop {
1950 enum ib_flow_spec_type type;
1954 union ib_flow_spec {
1959 struct ib_flow_spec_eth eth;
1960 struct ib_flow_spec_ib ib;
1961 struct ib_flow_spec_ipv4 ipv4;
1962 struct ib_flow_spec_tcp_udp tcp_udp;
1963 struct ib_flow_spec_ipv6 ipv6;
1964 struct ib_flow_spec_tunnel tunnel;
1965 struct ib_flow_spec_action_tag flow_tag;
1966 struct ib_flow_spec_action_drop drop;
1969 struct ib_flow_attr {
1970 enum ib_flow_attr_type type;
1976 /* Following are the optional layers according to user request
1977 * struct ib_flow_spec_xxx
1978 * struct ib_flow_spec_yyy
1984 struct ib_uobject *uobject;
1990 enum ib_process_mad_flags {
1991 IB_MAD_IGNORE_MKEY = 1,
1992 IB_MAD_IGNORE_BKEY = 2,
1993 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1996 enum ib_mad_result {
1997 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1998 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1999 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2000 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2003 struct ib_port_cache {
2005 struct ib_pkey_cache *pkey;
2006 struct ib_gid_table *gid;
2008 enum ib_port_state port_state;
2013 struct ib_event_handler event_handler;
2014 struct ib_port_cache *ports;
2019 struct ib_port_immutable {
2026 /* rdma netdev type - specifies protocol type */
2027 enum rdma_netdev_t {
2028 RDMA_NETDEV_OPA_VNIC,
2033 * struct rdma_netdev - rdma netdev
2034 * For cases where netstack interfacing is required.
2036 struct rdma_netdev {
2038 struct ib_device *hca;
2041 /* cleanup function must be specified */
2042 void (*free_rdma_netdev)(struct net_device *netdev);
2044 /* control functions */
2045 void (*set_id)(struct net_device *netdev, int id);
2047 int (*send)(struct net_device *dev, struct sk_buff *skb,
2048 struct ib_ah *address, u32 dqpn);
2050 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2051 union ib_gid *gid, u16 mlid,
2052 int set_qkey, u32 qkey);
2053 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2054 union ib_gid *gid, u16 mlid);
2057 struct ib_port_pkey_list {
2058 /* Lock to hold while modifying the list. */
2059 spinlock_t list_lock;
2060 struct list_head pkey_list;
2064 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2065 struct device *dma_device;
2067 char name[IB_DEVICE_NAME_MAX];
2069 struct list_head event_handler_list;
2070 spinlock_t event_handler_lock;
2072 spinlock_t client_data_lock;
2073 struct list_head core_list;
2074 /* Access to the client_data_list is protected by the client_data_lock
2075 * spinlock and the lists_rwsem read-write semaphore */
2076 struct list_head client_data_list;
2078 struct ib_cache cache;
2080 * port_immutable is indexed by port number
2082 struct ib_port_immutable *port_immutable;
2084 int num_comp_vectors;
2086 struct ib_port_pkey_list *port_pkey_list;
2088 struct iw_cm_verbs *iwcm;
2091 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2092 * driver initialized data. The struct is kfree()'ed by the sysfs
2093 * core when the device is removed. A lifespan of -1 in the return
2094 * struct tells the core to set a default lifespan.
2096 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2099 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2100 * @index - The index in the value array we wish to have updated, or
2101 * num_counters if we want all stats updated
2103 * < 0 - Error, no counters updated
2104 * index - Updated the single counter pointed to by index
2105 * num_counters - Updated all counters (will reset the timestamp
2106 * and prevent further calls for lifespan milliseconds)
2107 * Drivers are allowed to update all counters in leiu of just the
2108 * one given in index at their option
2110 int (*get_hw_stats)(struct ib_device *device,
2111 struct rdma_hw_stats *stats,
2112 u8 port, int index);
2113 int (*query_device)(struct ib_device *device,
2114 struct ib_device_attr *device_attr,
2115 struct ib_udata *udata);
2116 int (*query_port)(struct ib_device *device,
2118 struct ib_port_attr *port_attr);
2119 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2121 /* When calling get_netdev, the HW vendor's driver should return the
2122 * net device of device @device at port @port_num or NULL if such
2123 * a net device doesn't exist. The vendor driver should call dev_hold
2124 * on this net device. The HW vendor's device driver must guarantee
2125 * that this function returns NULL before the net device reaches
2126 * NETDEV_UNREGISTER_FINAL state.
2128 struct net_device *(*get_netdev)(struct ib_device *device,
2130 int (*query_gid)(struct ib_device *device,
2131 u8 port_num, int index,
2133 /* When calling add_gid, the HW vendor's driver should
2134 * add the gid of device @device at gid index @index of
2135 * port @port_num to be @gid. Meta-info of that gid (for example,
2136 * the network device related to this gid is available
2137 * at @attr. @context allows the HW vendor driver to store extra
2138 * information together with a GID entry. The HW vendor may allocate
2139 * memory to contain this information and store it in @context when a
2140 * new GID entry is written to. Params are consistent until the next
2141 * call of add_gid or delete_gid. The function should return 0 on
2142 * success or error otherwise. The function could be called
2143 * concurrently for different ports. This function is only called
2144 * when roce_gid_table is used.
2146 int (*add_gid)(struct ib_device *device,
2149 const union ib_gid *gid,
2150 const struct ib_gid_attr *attr,
2152 /* When calling del_gid, the HW vendor's driver should delete the
2153 * gid of device @device at gid index @index of port @port_num.
2154 * Upon the deletion of a GID entry, the HW vendor must free any
2155 * allocated memory. The caller will clear @context afterwards.
2156 * This function is only called when roce_gid_table is used.
2158 int (*del_gid)(struct ib_device *device,
2162 int (*query_pkey)(struct ib_device *device,
2163 u8 port_num, u16 index, u16 *pkey);
2164 int (*modify_device)(struct ib_device *device,
2165 int device_modify_mask,
2166 struct ib_device_modify *device_modify);
2167 int (*modify_port)(struct ib_device *device,
2168 u8 port_num, int port_modify_mask,
2169 struct ib_port_modify *port_modify);
2170 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2171 struct ib_udata *udata);
2172 int (*dealloc_ucontext)(struct ib_ucontext *context);
2173 int (*mmap)(struct ib_ucontext *context,
2174 struct vm_area_struct *vma);
2175 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2176 struct ib_ucontext *context,
2177 struct ib_udata *udata);
2178 int (*dealloc_pd)(struct ib_pd *pd);
2179 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2180 struct rdma_ah_attr *ah_attr,
2181 struct ib_udata *udata);
2182 int (*modify_ah)(struct ib_ah *ah,
2183 struct rdma_ah_attr *ah_attr);
2184 int (*query_ah)(struct ib_ah *ah,
2185 struct rdma_ah_attr *ah_attr);
2186 int (*destroy_ah)(struct ib_ah *ah);
2187 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2188 struct ib_srq_init_attr *srq_init_attr,
2189 struct ib_udata *udata);
2190 int (*modify_srq)(struct ib_srq *srq,
2191 struct ib_srq_attr *srq_attr,
2192 enum ib_srq_attr_mask srq_attr_mask,
2193 struct ib_udata *udata);
2194 int (*query_srq)(struct ib_srq *srq,
2195 struct ib_srq_attr *srq_attr);
2196 int (*destroy_srq)(struct ib_srq *srq);
2197 int (*post_srq_recv)(struct ib_srq *srq,
2198 struct ib_recv_wr *recv_wr,
2199 struct ib_recv_wr **bad_recv_wr);
2200 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2201 struct ib_qp_init_attr *qp_init_attr,
2202 struct ib_udata *udata);
2203 int (*modify_qp)(struct ib_qp *qp,
2204 struct ib_qp_attr *qp_attr,
2206 struct ib_udata *udata);
2207 int (*query_qp)(struct ib_qp *qp,
2208 struct ib_qp_attr *qp_attr,
2210 struct ib_qp_init_attr *qp_init_attr);
2211 int (*destroy_qp)(struct ib_qp *qp);
2212 int (*post_send)(struct ib_qp *qp,
2213 struct ib_send_wr *send_wr,
2214 struct ib_send_wr **bad_send_wr);
2215 int (*post_recv)(struct ib_qp *qp,
2216 struct ib_recv_wr *recv_wr,
2217 struct ib_recv_wr **bad_recv_wr);
2218 struct ib_cq * (*create_cq)(struct ib_device *device,
2219 const struct ib_cq_init_attr *attr,
2220 struct ib_ucontext *context,
2221 struct ib_udata *udata);
2222 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2224 int (*destroy_cq)(struct ib_cq *cq);
2225 int (*resize_cq)(struct ib_cq *cq, int cqe,
2226 struct ib_udata *udata);
2227 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2229 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2230 int (*req_notify_cq)(struct ib_cq *cq,
2231 enum ib_cq_notify_flags flags);
2232 int (*req_ncomp_notif)(struct ib_cq *cq,
2234 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2235 int mr_access_flags);
2236 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2237 u64 start, u64 length,
2239 int mr_access_flags,
2240 struct ib_udata *udata);
2241 int (*rereg_user_mr)(struct ib_mr *mr,
2243 u64 start, u64 length,
2245 int mr_access_flags,
2247 struct ib_udata *udata);
2248 int (*dereg_mr)(struct ib_mr *mr);
2249 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2250 enum ib_mr_type mr_type,
2252 int (*map_mr_sg)(struct ib_mr *mr,
2253 struct scatterlist *sg,
2255 unsigned int *sg_offset);
2256 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2257 enum ib_mw_type type,
2258 struct ib_udata *udata);
2259 int (*dealloc_mw)(struct ib_mw *mw);
2260 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2261 int mr_access_flags,
2262 struct ib_fmr_attr *fmr_attr);
2263 int (*map_phys_fmr)(struct ib_fmr *fmr,
2264 u64 *page_list, int list_len,
2266 int (*unmap_fmr)(struct list_head *fmr_list);
2267 int (*dealloc_fmr)(struct ib_fmr *fmr);
2268 int (*attach_mcast)(struct ib_qp *qp,
2271 int (*detach_mcast)(struct ib_qp *qp,
2274 int (*process_mad)(struct ib_device *device,
2275 int process_mad_flags,
2277 const struct ib_wc *in_wc,
2278 const struct ib_grh *in_grh,
2279 const struct ib_mad_hdr *in_mad,
2281 struct ib_mad_hdr *out_mad,
2282 size_t *out_mad_size,
2283 u16 *out_mad_pkey_index);
2284 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2285 struct ib_ucontext *ucontext,
2286 struct ib_udata *udata);
2287 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2288 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2292 int (*destroy_flow)(struct ib_flow *flow_id);
2293 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2294 struct ib_mr_status *mr_status);
2295 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2296 void (*drain_rq)(struct ib_qp *qp);
2297 void (*drain_sq)(struct ib_qp *qp);
2298 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2300 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2301 struct ifla_vf_info *ivf);
2302 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2303 struct ifla_vf_stats *stats);
2304 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2306 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2307 struct ib_wq_init_attr *init_attr,
2308 struct ib_udata *udata);
2309 int (*destroy_wq)(struct ib_wq *wq);
2310 int (*modify_wq)(struct ib_wq *wq,
2311 struct ib_wq_attr *attr,
2313 struct ib_udata *udata);
2314 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2315 struct ib_rwq_ind_table_init_attr *init_attr,
2316 struct ib_udata *udata);
2317 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2319 * rdma netdev operation
2321 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2322 * doesn't support the specified rdma netdev type.
2324 struct net_device *(*alloc_rdma_netdev)(
2325 struct ib_device *device,
2327 enum rdma_netdev_t type,
2329 unsigned char name_assign_type,
2330 void (*setup)(struct net_device *));
2332 struct module *owner;
2334 struct kobject *ports_parent;
2335 struct list_head port_list;
2338 IB_DEV_UNINITIALIZED,
2344 u64 uverbs_cmd_mask;
2345 u64 uverbs_ex_cmd_mask;
2347 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2353 struct ib_device_attr attrs;
2354 struct attribute_group *hw_stats_ag;
2355 struct rdma_hw_stats *hw_stats;
2357 #ifdef CONFIG_CGROUP_RDMA
2358 struct rdmacg_device cg_device;
2363 * Implementation details of the RDMA core, don't use in drivers
2365 struct rdma_restrack_root res;
2368 * The following mandatory functions are used only at device
2369 * registration. Keep functions such as these at the end of this
2370 * structure to avoid cache line misses when accessing struct ib_device
2373 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2374 void (*get_dev_fw_str)(struct ib_device *, char *str);
2375 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2378 struct uverbs_root_spec *specs_root;
2383 void (*add) (struct ib_device *);
2384 void (*remove)(struct ib_device *, void *client_data);
2386 /* Returns the net_dev belonging to this ib_client and matching the
2388 * @dev: An RDMA device that the net_dev use for communication.
2389 * @port: A physical port number on the RDMA device.
2390 * @pkey: P_Key that the net_dev uses if applicable.
2391 * @gid: A GID that the net_dev uses to communicate.
2392 * @addr: An IP address the net_dev is configured with.
2393 * @client_data: The device's client data set by ib_set_client_data().
2395 * An ib_client that implements a net_dev on top of RDMA devices
2396 * (such as IP over IB) should implement this callback, allowing the
2397 * rdma_cm module to find the right net_dev for a given request.
2399 * The caller is responsible for calling dev_put on the returned
2401 struct net_device *(*get_net_dev_by_params)(
2402 struct ib_device *dev,
2405 const union ib_gid *gid,
2406 const struct sockaddr *addr,
2408 struct list_head list;
2411 struct ib_device *ib_alloc_device(size_t size);
2412 void ib_dealloc_device(struct ib_device *device);
2414 void ib_get_device_fw_str(struct ib_device *device, char *str);
2416 int ib_register_device(struct ib_device *device,
2417 int (*port_callback)(struct ib_device *,
2418 u8, struct kobject *));
2419 void ib_unregister_device(struct ib_device *device);
2421 int ib_register_client (struct ib_client *client);
2422 void ib_unregister_client(struct ib_client *client);
2424 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2425 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2428 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2430 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2433 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2435 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2438 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2442 const void __user *p = udata->inbuf + offset;
2446 if (len > USHRT_MAX)
2449 buf = memdup_user(p, len);
2453 ret = !memchr_inv(buf, 0, len);
2459 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2460 * contains all required attributes and no attributes not allowed for
2461 * the given QP state transition.
2462 * @cur_state: Current QP state
2463 * @next_state: Next QP state
2465 * @mask: Mask of supplied QP attributes
2466 * @ll : link layer of port
2468 * This function is a helper function that a low-level driver's
2469 * modify_qp method can use to validate the consumer's input. It
2470 * checks that cur_state and next_state are valid QP states, that a
2471 * transition from cur_state to next_state is allowed by the IB spec,
2472 * and that the attribute mask supplied is allowed for the transition.
2474 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2475 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2476 enum rdma_link_layer ll);
2478 void ib_register_event_handler(struct ib_event_handler *event_handler);
2479 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2480 void ib_dispatch_event(struct ib_event *event);
2482 int ib_query_port(struct ib_device *device,
2483 u8 port_num, struct ib_port_attr *port_attr);
2485 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2489 * rdma_cap_ib_switch - Check if the device is IB switch
2490 * @device: Device to check
2492 * Device driver is responsible for setting is_switch bit on
2493 * in ib_device structure at init time.
2495 * Return: true if the device is IB switch.
2497 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2499 return device->is_switch;
2503 * rdma_start_port - Return the first valid port number for the device
2506 * @device: Device to be checked
2508 * Return start port number
2510 static inline u8 rdma_start_port(const struct ib_device *device)
2512 return rdma_cap_ib_switch(device) ? 0 : 1;
2516 * rdma_end_port - Return the last valid port number for the device
2519 * @device: Device to be checked
2521 * Return last port number
2523 static inline u8 rdma_end_port(const struct ib_device *device)
2525 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2528 static inline int rdma_is_port_valid(const struct ib_device *device,
2531 return (port >= rdma_start_port(device) &&
2532 port <= rdma_end_port(device));
2535 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2537 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2540 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2542 return device->port_immutable[port_num].core_cap_flags &
2543 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2546 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2548 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2551 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2553 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2556 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2558 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2561 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2563 return rdma_protocol_ib(device, port_num) ||
2564 rdma_protocol_roce(device, port_num);
2567 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2569 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2572 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2574 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2578 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2579 * Management Datagrams.
2580 * @device: Device to check
2581 * @port_num: Port number to check
2583 * Management Datagrams (MAD) are a required part of the InfiniBand
2584 * specification and are supported on all InfiniBand devices. A slightly
2585 * extended version are also supported on OPA interfaces.
2587 * Return: true if the port supports sending/receiving of MAD packets.
2589 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2591 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2595 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2596 * Management Datagrams.
2597 * @device: Device to check
2598 * @port_num: Port number to check
2600 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2601 * datagrams with their own versions. These OPA MADs share many but not all of
2602 * the characteristics of InfiniBand MADs.
2604 * OPA MADs differ in the following ways:
2606 * 1) MADs are variable size up to 2K
2607 * IBTA defined MADs remain fixed at 256 bytes
2608 * 2) OPA SMPs must carry valid PKeys
2609 * 3) OPA SMP packets are a different format
2611 * Return: true if the port supports OPA MAD packet formats.
2613 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2615 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2616 == RDMA_CORE_CAP_OPA_MAD;
2620 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2621 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2622 * @device: Device to check
2623 * @port_num: Port number to check
2625 * Each InfiniBand node is required to provide a Subnet Management Agent
2626 * that the subnet manager can access. Prior to the fabric being fully
2627 * configured by the subnet manager, the SMA is accessed via a well known
2628 * interface called the Subnet Management Interface (SMI). This interface
2629 * uses directed route packets to communicate with the SM to get around the
2630 * chicken and egg problem of the SM needing to know what's on the fabric
2631 * in order to configure the fabric, and needing to configure the fabric in
2632 * order to send packets to the devices on the fabric. These directed
2633 * route packets do not need the fabric fully configured in order to reach
2634 * their destination. The SMI is the only method allowed to send
2635 * directed route packets on an InfiniBand fabric.
2637 * Return: true if the port provides an SMI.
2639 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2641 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2645 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2646 * Communication Manager.
2647 * @device: Device to check
2648 * @port_num: Port number to check
2650 * The InfiniBand Communication Manager is one of many pre-defined General
2651 * Service Agents (GSA) that are accessed via the General Service
2652 * Interface (GSI). It's role is to facilitate establishment of connections
2653 * between nodes as well as other management related tasks for established
2656 * Return: true if the port supports an IB CM (this does not guarantee that
2657 * a CM is actually running however).
2659 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2661 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2665 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2666 * Communication Manager.
2667 * @device: Device to check
2668 * @port_num: Port number to check
2670 * Similar to above, but specific to iWARP connections which have a different
2671 * managment protocol than InfiniBand.
2673 * Return: true if the port supports an iWARP CM (this does not guarantee that
2674 * a CM is actually running however).
2676 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2678 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2682 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2683 * Subnet Administration.
2684 * @device: Device to check
2685 * @port_num: Port number to check
2687 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2688 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2689 * fabrics, devices should resolve routes to other hosts by contacting the
2690 * SA to query the proper route.
2692 * Return: true if the port should act as a client to the fabric Subnet
2693 * Administration interface. This does not imply that the SA service is
2696 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2698 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2702 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2704 * @device: Device to check
2705 * @port_num: Port number to check
2707 * InfiniBand multicast registration is more complex than normal IPv4 or
2708 * IPv6 multicast registration. Each Host Channel Adapter must register
2709 * with the Subnet Manager when it wishes to join a multicast group. It
2710 * should do so only once regardless of how many queue pairs it subscribes
2711 * to this group. And it should leave the group only after all queue pairs
2712 * attached to the group have been detached.
2714 * Return: true if the port must undertake the additional adminstrative
2715 * overhead of registering/unregistering with the SM and tracking of the
2716 * total number of queue pairs attached to the multicast group.
2718 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2720 return rdma_cap_ib_sa(device, port_num);
2724 * rdma_cap_af_ib - Check if the port of device has the capability
2725 * Native Infiniband Address.
2726 * @device: Device to check
2727 * @port_num: Port number to check
2729 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2730 * GID. RoCE uses a different mechanism, but still generates a GID via
2731 * a prescribed mechanism and port specific data.
2733 * Return: true if the port uses a GID address to identify devices on the
2736 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2738 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2742 * rdma_cap_eth_ah - Check if the port of device has the capability
2743 * Ethernet Address Handle.
2744 * @device: Device to check
2745 * @port_num: Port number to check
2747 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2748 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2749 * port. Normally, packet headers are generated by the sending host
2750 * adapter, but when sending connectionless datagrams, we must manually
2751 * inject the proper headers for the fabric we are communicating over.
2753 * Return: true if we are running as a RoCE port and must force the
2754 * addition of a Global Route Header built from our Ethernet Address
2755 * Handle into our header list for connectionless packets.
2757 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2759 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2763 * rdma_cap_opa_ah - Check if the port of device supports
2764 * OPA Address handles
2765 * @device: Device to check
2766 * @port_num: Port number to check
2768 * Return: true if we are running on an OPA device which supports
2769 * the extended OPA addressing.
2771 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2773 return (device->port_immutable[port_num].core_cap_flags &
2774 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2778 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2781 * @port_num: Port number
2783 * This MAD size includes the MAD headers and MAD payload. No other headers
2786 * Return the max MAD size required by the Port. Will return 0 if the port
2787 * does not support MADs
2789 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2791 return device->port_immutable[port_num].max_mad_size;
2795 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2796 * @device: Device to check
2797 * @port_num: Port number to check
2799 * RoCE GID table mechanism manages the various GIDs for a device.
2801 * NOTE: if allocating the port's GID table has failed, this call will still
2802 * return true, but any RoCE GID table API will fail.
2804 * Return: true if the port uses RoCE GID table mechanism in order to manage
2807 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2810 return rdma_protocol_roce(device, port_num) &&
2811 device->add_gid && device->del_gid;
2815 * Check if the device supports READ W/ INVALIDATE.
2817 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2820 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2821 * has support for it yet.
2823 return rdma_protocol_iwarp(dev, port_num);
2826 int ib_query_gid(struct ib_device *device,
2827 u8 port_num, int index, union ib_gid *gid,
2828 struct ib_gid_attr *attr);
2830 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2832 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2833 struct ifla_vf_info *info);
2834 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2835 struct ifla_vf_stats *stats);
2836 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2839 int ib_query_pkey(struct ib_device *device,
2840 u8 port_num, u16 index, u16 *pkey);
2842 int ib_modify_device(struct ib_device *device,
2843 int device_modify_mask,
2844 struct ib_device_modify *device_modify);
2846 int ib_modify_port(struct ib_device *device,
2847 u8 port_num, int port_modify_mask,
2848 struct ib_port_modify *port_modify);
2850 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2851 struct net_device *ndev, u8 *port_num, u16 *index);
2853 int ib_find_pkey(struct ib_device *device,
2854 u8 port_num, u16 pkey, u16 *index);
2858 * Create a memory registration for all memory in the system and place
2859 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2860 * ULPs to avoid the overhead of dynamic MRs.
2862 * This flag is generally considered unsafe and must only be used in
2863 * extremly trusted environments. Every use of it will log a warning
2864 * in the kernel log.
2866 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2869 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2870 const char *caller);
2871 #define ib_alloc_pd(device, flags) \
2872 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
2873 void ib_dealloc_pd(struct ib_pd *pd);
2876 * rdma_create_ah - Creates an address handle for the given address vector.
2877 * @pd: The protection domain associated with the address handle.
2878 * @ah_attr: The attributes of the address vector.
2880 * The address handle is used to reference a local or global destination
2881 * in all UD QP post sends.
2883 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2886 * rdma_create_user_ah - Creates an address handle for the given address vector.
2887 * It resolves destination mac address for ah attribute of RoCE type.
2888 * @pd: The protection domain associated with the address handle.
2889 * @ah_attr: The attributes of the address vector.
2890 * @udata: pointer to user's input output buffer information need by
2893 * It returns 0 on success and returns appropriate error code on error.
2894 * The address handle is used to reference a local or global destination
2895 * in all UD QP post sends.
2897 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
2898 struct rdma_ah_attr *ah_attr,
2899 struct ib_udata *udata);
2901 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2903 * @hdr: the L3 header to parse
2904 * @net_type: type of header to parse
2905 * @sgid: place to store source gid
2906 * @dgid: place to store destination gid
2908 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2909 enum rdma_network_type net_type,
2910 union ib_gid *sgid, union ib_gid *dgid);
2913 * ib_get_rdma_header_version - Get the header version
2914 * @hdr: the L3 header to parse
2916 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2919 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
2921 * @device: Device on which the received message arrived.
2922 * @port_num: Port on which the received message arrived.
2923 * @wc: Work completion associated with the received message.
2924 * @grh: References the received global route header. This parameter is
2925 * ignored unless the work completion indicates that the GRH is valid.
2926 * @ah_attr: Returned attributes that can be used when creating an address
2927 * handle for replying to the message.
2929 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
2930 const struct ib_wc *wc, const struct ib_grh *grh,
2931 struct rdma_ah_attr *ah_attr);
2934 * ib_create_ah_from_wc - Creates an address handle associated with the
2935 * sender of the specified work completion.
2936 * @pd: The protection domain associated with the address handle.
2937 * @wc: Work completion information associated with a received message.
2938 * @grh: References the received global route header. This parameter is
2939 * ignored unless the work completion indicates that the GRH is valid.
2940 * @port_num: The outbound port number to associate with the address.
2942 * The address handle is used to reference a local or global destination
2943 * in all UD QP post sends.
2945 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2946 const struct ib_grh *grh, u8 port_num);
2949 * rdma_modify_ah - Modifies the address vector associated with an address
2951 * @ah: The address handle to modify.
2952 * @ah_attr: The new address vector attributes to associate with the
2955 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2958 * rdma_query_ah - Queries the address vector associated with an address
2960 * @ah: The address handle to query.
2961 * @ah_attr: The address vector attributes associated with the address
2964 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2967 * rdma_destroy_ah - Destroys an address handle.
2968 * @ah: The address handle to destroy.
2970 int rdma_destroy_ah(struct ib_ah *ah);
2973 * ib_create_srq - Creates a SRQ associated with the specified protection
2975 * @pd: The protection domain associated with the SRQ.
2976 * @srq_init_attr: A list of initial attributes required to create the
2977 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2978 * the actual capabilities of the created SRQ.
2980 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2981 * requested size of the SRQ, and set to the actual values allocated
2982 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2983 * will always be at least as large as the requested values.
2985 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2986 struct ib_srq_init_attr *srq_init_attr);
2989 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2990 * @srq: The SRQ to modify.
2991 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2992 * the current values of selected SRQ attributes are returned.
2993 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2994 * are being modified.
2996 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2997 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2998 * the number of receives queued drops below the limit.
3000 int ib_modify_srq(struct ib_srq *srq,
3001 struct ib_srq_attr *srq_attr,
3002 enum ib_srq_attr_mask srq_attr_mask);
3005 * ib_query_srq - Returns the attribute list and current values for the
3007 * @srq: The SRQ to query.
3008 * @srq_attr: The attributes of the specified SRQ.
3010 int ib_query_srq(struct ib_srq *srq,
3011 struct ib_srq_attr *srq_attr);
3014 * ib_destroy_srq - Destroys the specified SRQ.
3015 * @srq: The SRQ to destroy.
3017 int ib_destroy_srq(struct ib_srq *srq);
3020 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3021 * @srq: The SRQ to post the work request on.
3022 * @recv_wr: A list of work requests to post on the receive queue.
3023 * @bad_recv_wr: On an immediate failure, this parameter will reference
3024 * the work request that failed to be posted on the QP.
3026 static inline int ib_post_srq_recv(struct ib_srq *srq,
3027 struct ib_recv_wr *recv_wr,
3028 struct ib_recv_wr **bad_recv_wr)
3030 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3034 * ib_create_qp - Creates a QP associated with the specified protection
3036 * @pd: The protection domain associated with the QP.
3037 * @qp_init_attr: A list of initial attributes required to create the
3038 * QP. If QP creation succeeds, then the attributes are updated to
3039 * the actual capabilities of the created QP.
3041 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3042 struct ib_qp_init_attr *qp_init_attr);
3045 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3046 * @qp: The QP to modify.
3047 * @attr: On input, specifies the QP attributes to modify. On output,
3048 * the current values of selected QP attributes are returned.
3049 * @attr_mask: A bit-mask used to specify which attributes of the QP
3050 * are being modified.
3051 * @udata: pointer to user's input output buffer information
3052 * are being modified.
3053 * It returns 0 on success and returns appropriate error code on error.
3055 int ib_modify_qp_with_udata(struct ib_qp *qp,
3056 struct ib_qp_attr *attr,
3058 struct ib_udata *udata);
3061 * ib_modify_qp - Modifies the attributes for the specified QP and then
3062 * transitions the QP to the given state.
3063 * @qp: The QP to modify.
3064 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3065 * the current values of selected QP attributes are returned.
3066 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3067 * are being modified.
3069 int ib_modify_qp(struct ib_qp *qp,
3070 struct ib_qp_attr *qp_attr,
3074 * ib_query_qp - Returns the attribute list and current values for the
3076 * @qp: The QP to query.
3077 * @qp_attr: The attributes of the specified QP.
3078 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3079 * @qp_init_attr: Additional attributes of the selected QP.
3081 * The qp_attr_mask may be used to limit the query to gathering only the
3082 * selected attributes.
3084 int ib_query_qp(struct ib_qp *qp,
3085 struct ib_qp_attr *qp_attr,
3087 struct ib_qp_init_attr *qp_init_attr);
3090 * ib_destroy_qp - Destroys the specified QP.
3091 * @qp: The QP to destroy.
3093 int ib_destroy_qp(struct ib_qp *qp);
3096 * ib_open_qp - Obtain a reference to an existing sharable QP.
3097 * @xrcd - XRC domain
3098 * @qp_open_attr: Attributes identifying the QP to open.
3100 * Returns a reference to a sharable QP.
3102 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3103 struct ib_qp_open_attr *qp_open_attr);
3106 * ib_close_qp - Release an external reference to a QP.
3107 * @qp: The QP handle to release
3109 * The opened QP handle is released by the caller. The underlying
3110 * shared QP is not destroyed until all internal references are released.
3112 int ib_close_qp(struct ib_qp *qp);
3115 * ib_post_send - Posts a list of work requests to the send queue of
3117 * @qp: The QP to post the work request on.
3118 * @send_wr: A list of work requests to post on the send queue.
3119 * @bad_send_wr: On an immediate failure, this parameter will reference
3120 * the work request that failed to be posted on the QP.
3122 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3123 * error is returned, the QP state shall not be affected,
3124 * ib_post_send() will return an immediate error after queueing any
3125 * earlier work requests in the list.
3127 static inline int ib_post_send(struct ib_qp *qp,
3128 struct ib_send_wr *send_wr,
3129 struct ib_send_wr **bad_send_wr)
3131 return qp->device->post_send(qp, send_wr, bad_send_wr);
3135 * ib_post_recv - Posts a list of work requests to the receive queue of
3137 * @qp: The QP to post the work request on.
3138 * @recv_wr: A list of work requests to post on the receive queue.
3139 * @bad_recv_wr: On an immediate failure, this parameter will reference
3140 * the work request that failed to be posted on the QP.
3142 static inline int ib_post_recv(struct ib_qp *qp,
3143 struct ib_recv_wr *recv_wr,
3144 struct ib_recv_wr **bad_recv_wr)
3146 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3149 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3150 int nr_cqe, int comp_vector,
3151 enum ib_poll_context poll_ctx, const char *caller);
3152 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3153 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3155 void ib_free_cq(struct ib_cq *cq);
3156 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3159 * ib_create_cq - Creates a CQ on the specified device.
3160 * @device: The device on which to create the CQ.
3161 * @comp_handler: A user-specified callback that is invoked when a
3162 * completion event occurs on the CQ.
3163 * @event_handler: A user-specified callback that is invoked when an
3164 * asynchronous event not associated with a completion occurs on the CQ.
3165 * @cq_context: Context associated with the CQ returned to the user via
3166 * the associated completion and event handlers.
3167 * @cq_attr: The attributes the CQ should be created upon.
3169 * Users can examine the cq structure to determine the actual CQ size.
3171 struct ib_cq *ib_create_cq(struct ib_device *device,
3172 ib_comp_handler comp_handler,
3173 void (*event_handler)(struct ib_event *, void *),
3175 const struct ib_cq_init_attr *cq_attr);
3178 * ib_resize_cq - Modifies the capacity of the CQ.
3179 * @cq: The CQ to resize.
3180 * @cqe: The minimum size of the CQ.
3182 * Users can examine the cq structure to determine the actual CQ size.
3184 int ib_resize_cq(struct ib_cq *cq, int cqe);
3187 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3188 * @cq: The CQ to modify.
3189 * @cq_count: number of CQEs that will trigger an event
3190 * @cq_period: max period of time in usec before triggering an event
3193 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3196 * ib_destroy_cq - Destroys the specified CQ.
3197 * @cq: The CQ to destroy.
3199 int ib_destroy_cq(struct ib_cq *cq);
3202 * ib_poll_cq - poll a CQ for completion(s)
3203 * @cq:the CQ being polled
3204 * @num_entries:maximum number of completions to return
3205 * @wc:array of at least @num_entries &struct ib_wc where completions
3208 * Poll a CQ for (possibly multiple) completions. If the return value
3209 * is < 0, an error occurred. If the return value is >= 0, it is the
3210 * number of completions returned. If the return value is
3211 * non-negative and < num_entries, then the CQ was emptied.
3213 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3216 return cq->device->poll_cq(cq, num_entries, wc);
3220 * ib_peek_cq - Returns the number of unreaped completions currently
3221 * on the specified CQ.
3222 * @cq: The CQ to peek.
3223 * @wc_cnt: A minimum number of unreaped completions to check for.
3225 * If the number of unreaped completions is greater than or equal to wc_cnt,
3226 * this function returns wc_cnt, otherwise, it returns the actual number of
3227 * unreaped completions.
3229 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3232 * ib_req_notify_cq - Request completion notification on a CQ.
3233 * @cq: The CQ to generate an event for.
3235 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3236 * to request an event on the next solicited event or next work
3237 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3238 * may also be |ed in to request a hint about missed events, as
3242 * < 0 means an error occurred while requesting notification
3243 * == 0 means notification was requested successfully, and if
3244 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3245 * were missed and it is safe to wait for another event. In
3246 * this case is it guaranteed that any work completions added
3247 * to the CQ since the last CQ poll will trigger a completion
3248 * notification event.
3249 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3250 * in. It means that the consumer must poll the CQ again to
3251 * make sure it is empty to avoid missing an event because of a
3252 * race between requesting notification and an entry being
3253 * added to the CQ. This return value means it is possible
3254 * (but not guaranteed) that a work completion has been added
3255 * to the CQ since the last poll without triggering a
3256 * completion notification event.
3258 static inline int ib_req_notify_cq(struct ib_cq *cq,
3259 enum ib_cq_notify_flags flags)
3261 return cq->device->req_notify_cq(cq, flags);
3265 * ib_req_ncomp_notif - Request completion notification when there are
3266 * at least the specified number of unreaped completions on the CQ.
3267 * @cq: The CQ to generate an event for.
3268 * @wc_cnt: The number of unreaped completions that should be on the
3269 * CQ before an event is generated.
3271 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3273 return cq->device->req_ncomp_notif ?
3274 cq->device->req_ncomp_notif(cq, wc_cnt) :
3279 * ib_dma_mapping_error - check a DMA addr for error
3280 * @dev: The device for which the dma_addr was created
3281 * @dma_addr: The DMA address to check
3283 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3285 return dma_mapping_error(dev->dma_device, dma_addr);
3289 * ib_dma_map_single - Map a kernel virtual address to DMA address
3290 * @dev: The device for which the dma_addr is to be created
3291 * @cpu_addr: The kernel virtual address
3292 * @size: The size of the region in bytes
3293 * @direction: The direction of the DMA
3295 static inline u64 ib_dma_map_single(struct ib_device *dev,
3296 void *cpu_addr, size_t size,
3297 enum dma_data_direction direction)
3299 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3303 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3304 * @dev: The device for which the DMA address was created
3305 * @addr: The DMA address
3306 * @size: The size of the region in bytes
3307 * @direction: The direction of the DMA
3309 static inline void ib_dma_unmap_single(struct ib_device *dev,
3310 u64 addr, size_t size,
3311 enum dma_data_direction direction)
3313 dma_unmap_single(dev->dma_device, addr, size, direction);
3317 * ib_dma_map_page - Map a physical page to DMA address
3318 * @dev: The device for which the dma_addr is to be created
3319 * @page: The page to be mapped
3320 * @offset: The offset within the page
3321 * @size: The size of the region in bytes
3322 * @direction: The direction of the DMA
3324 static inline u64 ib_dma_map_page(struct ib_device *dev,
3326 unsigned long offset,
3328 enum dma_data_direction direction)
3330 return dma_map_page(dev->dma_device, page, offset, size, direction);
3334 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3335 * @dev: The device for which the DMA address was created
3336 * @addr: The DMA address
3337 * @size: The size of the region in bytes
3338 * @direction: The direction of the DMA
3340 static inline void ib_dma_unmap_page(struct ib_device *dev,
3341 u64 addr, size_t size,
3342 enum dma_data_direction direction)
3344 dma_unmap_page(dev->dma_device, addr, size, direction);
3348 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3349 * @dev: The device for which the DMA addresses are to be created
3350 * @sg: The array of scatter/gather entries
3351 * @nents: The number of scatter/gather entries
3352 * @direction: The direction of the DMA
3354 static inline int ib_dma_map_sg(struct ib_device *dev,
3355 struct scatterlist *sg, int nents,
3356 enum dma_data_direction direction)
3358 return dma_map_sg(dev->dma_device, sg, nents, direction);
3362 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3363 * @dev: The device for which the DMA addresses were created
3364 * @sg: The array of scatter/gather entries
3365 * @nents: The number of scatter/gather entries
3366 * @direction: The direction of the DMA
3368 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3369 struct scatterlist *sg, int nents,
3370 enum dma_data_direction direction)
3372 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3375 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3376 struct scatterlist *sg, int nents,
3377 enum dma_data_direction direction,
3378 unsigned long dma_attrs)
3380 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3384 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3385 struct scatterlist *sg, int nents,
3386 enum dma_data_direction direction,
3387 unsigned long dma_attrs)
3389 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3392 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3393 * @dev: The device for which the DMA addresses were created
3394 * @sg: The scatter/gather entry
3396 * Note: this function is obsolete. To do: change all occurrences of
3397 * ib_sg_dma_address() into sg_dma_address().
3399 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3400 struct scatterlist *sg)
3402 return sg_dma_address(sg);
3406 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3407 * @dev: The device for which the DMA addresses were created
3408 * @sg: The scatter/gather entry
3410 * Note: this function is obsolete. To do: change all occurrences of
3411 * ib_sg_dma_len() into sg_dma_len().
3413 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3414 struct scatterlist *sg)
3416 return sg_dma_len(sg);
3420 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3421 * @dev: The device for which the DMA address was created
3422 * @addr: The DMA address
3423 * @size: The size of the region in bytes
3424 * @dir: The direction of the DMA
3426 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3429 enum dma_data_direction dir)
3431 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3435 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3436 * @dev: The device for which the DMA address was created
3437 * @addr: The DMA address
3438 * @size: The size of the region in bytes
3439 * @dir: The direction of the DMA
3441 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3444 enum dma_data_direction dir)
3446 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3450 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3451 * @dev: The device for which the DMA address is requested
3452 * @size: The size of the region to allocate in bytes
3453 * @dma_handle: A pointer for returning the DMA address of the region
3454 * @flag: memory allocator flags
3456 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3458 dma_addr_t *dma_handle,
3461 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3465 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3466 * @dev: The device for which the DMA addresses were allocated
3467 * @size: The size of the region
3468 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3469 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3471 static inline void ib_dma_free_coherent(struct ib_device *dev,
3472 size_t size, void *cpu_addr,
3473 dma_addr_t dma_handle)
3475 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3479 * ib_dereg_mr - Deregisters a memory region and removes it from the
3480 * HCA translation table.
3481 * @mr: The memory region to deregister.
3483 * This function can fail, if the memory region has memory windows bound to it.
3485 int ib_dereg_mr(struct ib_mr *mr);
3487 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3488 enum ib_mr_type mr_type,
3492 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3494 * @mr - struct ib_mr pointer to be updated.
3495 * @newkey - new key to be used.
3497 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3499 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3500 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3504 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3505 * for calculating a new rkey for type 2 memory windows.
3506 * @rkey - the rkey to increment.
3508 static inline u32 ib_inc_rkey(u32 rkey)
3510 const u32 mask = 0x000000ff;
3511 return ((rkey + 1) & mask) | (rkey & ~mask);
3515 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3516 * @pd: The protection domain associated with the unmapped region.
3517 * @mr_access_flags: Specifies the memory access rights.
3518 * @fmr_attr: Attributes of the unmapped region.
3520 * A fast memory region must be mapped before it can be used as part of
3523 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3524 int mr_access_flags,
3525 struct ib_fmr_attr *fmr_attr);
3528 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3529 * @fmr: The fast memory region to associate with the pages.
3530 * @page_list: An array of physical pages to map to the fast memory region.
3531 * @list_len: The number of pages in page_list.
3532 * @iova: The I/O virtual address to use with the mapped region.
3534 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3535 u64 *page_list, int list_len,
3538 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3542 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3543 * @fmr_list: A linked list of fast memory regions to unmap.
3545 int ib_unmap_fmr(struct list_head *fmr_list);
3548 * ib_dealloc_fmr - Deallocates a fast memory region.
3549 * @fmr: The fast memory region to deallocate.
3551 int ib_dealloc_fmr(struct ib_fmr *fmr);
3554 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3555 * @qp: QP to attach to the multicast group. The QP must be type
3557 * @gid: Multicast group GID.
3558 * @lid: Multicast group LID in host byte order.
3560 * In order to send and receive multicast packets, subnet
3561 * administration must have created the multicast group and configured
3562 * the fabric appropriately. The port associated with the specified
3563 * QP must also be a member of the multicast group.
3565 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3568 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3569 * @qp: QP to detach from the multicast group.
3570 * @gid: Multicast group GID.
3571 * @lid: Multicast group LID in host byte order.
3573 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3576 * ib_alloc_xrcd - Allocates an XRC domain.
3577 * @device: The device on which to allocate the XRC domain.
3578 * @caller: Module name for kernel consumers
3580 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3581 #define ib_alloc_xrcd(device) \
3582 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3585 * ib_dealloc_xrcd - Deallocates an XRC domain.
3586 * @xrcd: The XRC domain to deallocate.
3588 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3590 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3591 struct ib_flow_attr *flow_attr, int domain);
3592 int ib_destroy_flow(struct ib_flow *flow_id);
3594 static inline int ib_check_mr_access(int flags)
3597 * Local write permission is required if remote write or
3598 * remote atomic permission is also requested.
3600 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3601 !(flags & IB_ACCESS_LOCAL_WRITE))
3608 * ib_check_mr_status: lightweight check of MR status.
3609 * This routine may provide status checks on a selected
3610 * ib_mr. first use is for signature status check.
3612 * @mr: A memory region.
3613 * @check_mask: Bitmask of which checks to perform from
3614 * ib_mr_status_check enumeration.
3615 * @mr_status: The container of relevant status checks.
3616 * failed checks will be indicated in the status bitmask
3617 * and the relevant info shall be in the error item.
3619 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3620 struct ib_mr_status *mr_status);
3622 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3623 u16 pkey, const union ib_gid *gid,
3624 const struct sockaddr *addr);
3625 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3626 struct ib_wq_init_attr *init_attr);
3627 int ib_destroy_wq(struct ib_wq *wq);
3628 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3630 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3631 struct ib_rwq_ind_table_init_attr*
3632 wq_ind_table_init_attr);
3633 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3635 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3636 unsigned int *sg_offset, unsigned int page_size);
3639 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3640 unsigned int *sg_offset, unsigned int page_size)
3644 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3650 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3651 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3653 void ib_drain_rq(struct ib_qp *qp);
3654 void ib_drain_sq(struct ib_qp *qp);
3655 void ib_drain_qp(struct ib_qp *qp);
3657 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3659 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3661 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3662 return attr->roce.dmac;
3666 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3668 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3669 attr->ib.dlid = (u16)dlid;
3670 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3671 attr->opa.dlid = dlid;
3674 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3676 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3677 return attr->ib.dlid;
3678 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3679 return attr->opa.dlid;
3683 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3688 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3693 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3696 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3697 attr->ib.src_path_bits = src_path_bits;
3698 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3699 attr->opa.src_path_bits = src_path_bits;
3702 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3704 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3705 return attr->ib.src_path_bits;
3706 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3707 return attr->opa.src_path_bits;
3711 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3714 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3715 attr->opa.make_grd = make_grd;
3718 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3720 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3721 return attr->opa.make_grd;
3725 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3727 attr->port_num = port_num;
3730 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3732 return attr->port_num;
3735 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3738 attr->static_rate = static_rate;
3741 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3743 return attr->static_rate;
3746 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3747 enum ib_ah_flags flag)
3749 attr->ah_flags = flag;
3752 static inline enum ib_ah_flags
3753 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3755 return attr->ah_flags;
3758 static inline const struct ib_global_route
3759 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3764 /*To retrieve and modify the grh */
3765 static inline struct ib_global_route
3766 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3771 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3773 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3775 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3778 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3781 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3783 grh->dgid.global.subnet_prefix = prefix;
3786 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3789 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3791 grh->dgid.global.interface_id = if_id;
3794 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3795 union ib_gid *dgid, u32 flow_label,
3796 u8 sgid_index, u8 hop_limit,
3799 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3801 attr->ah_flags = IB_AH_GRH;
3804 grh->flow_label = flow_label;
3805 grh->sgid_index = sgid_index;
3806 grh->hop_limit = hop_limit;
3807 grh->traffic_class = traffic_class;
3811 * rdma_ah_find_type - Return address handle type.
3813 * @dev: Device to be checked
3814 * @port_num: Port number
3816 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3819 if (rdma_protocol_roce(dev, port_num))
3820 return RDMA_AH_ATTR_TYPE_ROCE;
3821 if (rdma_protocol_ib(dev, port_num)) {
3822 if (rdma_cap_opa_ah(dev, port_num))
3823 return RDMA_AH_ATTR_TYPE_OPA;
3824 return RDMA_AH_ATTR_TYPE_IB;
3827 return RDMA_AH_ATTR_TYPE_UNDEFINED;
3831 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3832 * In the current implementation the only way to get
3833 * get the 32bit lid is from other sources for OPA.
3834 * For IB, lids will always be 16bits so cast the
3835 * value accordingly.
3839 static inline u16 ib_lid_cpu16(u32 lid)
3841 WARN_ON_ONCE(lid & 0xFFFF0000);
3846 * ib_lid_be16 - Return lid in 16bit BE encoding.
3850 static inline __be16 ib_lid_be16(u32 lid)
3852 WARN_ON_ONCE(lid & 0xFFFF0000);
3853 return cpu_to_be16((u16)lid);
3857 * ib_get_vector_affinity - Get the affinity mappings of a given completion
3859 * @device: the rdma device
3860 * @comp_vector: index of completion vector
3862 * Returns NULL on failure, otherwise a corresponding cpu map of the
3863 * completion vector (returns all-cpus map if the device driver doesn't
3864 * implement get_vector_affinity).
3866 static inline const struct cpumask *
3867 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3869 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3870 !device->get_vector_affinity)
3873 return device->get_vector_affinity(device, comp_vector);
3878 * rdma_roce_rescan_device - Rescan all of the network devices in the system
3879 * and add their gids, as needed, to the relevant RoCE devices.
3881 * @device: the rdma device
3883 void rdma_roce_rescan_device(struct ib_device *ibdev);
3885 #endif /* IB_VERBS_H */