2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
69 extern struct workqueue_struct *ib_wq;
70 extern struct workqueue_struct *ib_comp_wq;
80 extern union ib_gid zgid;
83 /* If link layer is Ethernet, this is RoCE V1 */
86 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
90 #define ROCE_V2_UDP_DPORT 4791
92 enum ib_gid_type gid_type;
93 struct net_device *ndev;
97 /* IB values map to NodeInfo:NodeType. */
107 /* set the local administered indication */
108 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
111 enum rdma_transport_type {
113 RDMA_TRANSPORT_IWARP,
114 RDMA_TRANSPORT_USNIC,
115 RDMA_TRANSPORT_USNIC_UDP
118 enum rdma_protocol_type {
122 RDMA_PROTOCOL_USNIC_UDP
125 __attribute_const__ enum rdma_transport_type
126 rdma_node_get_transport(enum rdma_node_type node_type);
128 enum rdma_network_type {
130 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
135 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
137 if (network_type == RDMA_NETWORK_IPV4 ||
138 network_type == RDMA_NETWORK_IPV6)
139 return IB_GID_TYPE_ROCE_UDP_ENCAP;
141 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
142 return IB_GID_TYPE_IB;
145 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
148 if (gid_type == IB_GID_TYPE_IB)
149 return RDMA_NETWORK_IB;
151 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
152 return RDMA_NETWORK_IPV4;
154 return RDMA_NETWORK_IPV6;
157 enum rdma_link_layer {
158 IB_LINK_LAYER_UNSPECIFIED,
159 IB_LINK_LAYER_INFINIBAND,
160 IB_LINK_LAYER_ETHERNET,
163 enum ib_device_cap_flags {
164 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
165 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
166 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
167 IB_DEVICE_RAW_MULTI = (1 << 3),
168 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
169 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
170 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
171 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
172 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
173 /* Not in use, former INIT_TYPE = (1 << 9),*/
174 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
175 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
176 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
177 IB_DEVICE_SRQ_RESIZE = (1 << 13),
178 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
181 * This device supports a per-device lkey or stag that can be
182 * used without performing a memory registration for the local
183 * memory. Note that ULPs should never check this flag, but
184 * instead of use the local_dma_lkey flag in the ib_pd structure,
185 * which will always contain a usable lkey.
187 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
188 /* Reserved, old SEND_W_INV = (1 << 16),*/
189 IB_DEVICE_MEM_WINDOW = (1 << 17),
191 * Devices should set IB_DEVICE_UD_IP_SUM if they support
192 * insertion of UDP and TCP checksum on outgoing UD IPoIB
193 * messages and can verify the validity of checksum for
194 * incoming messages. Setting this flag implies that the
195 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
197 IB_DEVICE_UD_IP_CSUM = (1 << 18),
198 IB_DEVICE_UD_TSO = (1 << 19),
199 IB_DEVICE_XRC = (1 << 20),
202 * This device supports the IB "base memory management extension",
203 * which includes support for fast registrations (IB_WR_REG_MR,
204 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
205 * also be set by any iWarp device which must support FRs to comply
206 * to the iWarp verbs spec. iWarp devices also support the
207 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
210 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
211 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
212 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
213 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
214 IB_DEVICE_RC_IP_CSUM = (1 << 25),
215 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
216 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
218 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
219 * support execution of WQEs that involve synchronization
220 * of I/O operations with single completion queue managed
223 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
224 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
225 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
226 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
227 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
228 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
229 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
230 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
231 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
232 /* The device supports padding incoming writes to cacheline. */
233 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
236 enum ib_signature_prot_cap {
237 IB_PROT_T10DIF_TYPE_1 = 1,
238 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
239 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
242 enum ib_signature_guard_cap {
243 IB_GUARD_T10DIF_CRC = 1,
244 IB_GUARD_T10DIF_CSUM = 1 << 1,
253 enum ib_odp_general_cap_bits {
254 IB_ODP_SUPPORT = 1 << 0,
255 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
258 enum ib_odp_transport_cap_bits {
259 IB_ODP_SUPPORT_SEND = 1 << 0,
260 IB_ODP_SUPPORT_RECV = 1 << 1,
261 IB_ODP_SUPPORT_WRITE = 1 << 2,
262 IB_ODP_SUPPORT_READ = 1 << 3,
263 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
267 uint64_t general_caps;
269 uint32_t rc_odp_caps;
270 uint32_t uc_odp_caps;
271 uint32_t ud_odp_caps;
272 } per_transport_caps;
276 /* Corresponding bit will be set if qp type from
277 * 'enum ib_qp_type' is supported, e.g.
278 * supported_qpts |= 1 << IB_QPT_UD
281 u32 max_rwq_indirection_tables;
282 u32 max_rwq_indirection_table_size;
285 enum ib_tm_cap_flags {
286 /* Support tag matching on RC transport */
287 IB_TM_CAP_RC = 1 << 0,
291 /* Max size of RNDV header */
292 u32 max_rndv_hdr_size;
293 /* Max number of entries in tag matching list */
295 /* From enum ib_tm_cap_flags */
297 /* Max number of outstanding list operations */
299 /* Max number of SGE in tag matching entry */
303 struct ib_cq_init_attr {
309 enum ib_cq_attr_mask {
310 IB_CQ_MODERATE = 1 << 0,
314 u16 max_cq_moderation_count;
315 u16 max_cq_moderation_period;
318 struct ib_device_attr {
320 __be64 sys_image_guid;
328 u64 device_cap_flags;
338 int max_qp_init_rd_atom;
339 int max_ee_init_rd_atom;
340 enum ib_atomic_cap atomic_cap;
341 enum ib_atomic_cap masked_atomic_cap;
348 int max_mcast_qp_attach;
349 int max_total_mcast_qp_attach;
356 unsigned int max_fast_reg_page_list_len;
358 u8 local_ca_ack_delay;
361 struct ib_odp_caps odp_caps;
362 uint64_t timestamp_mask;
363 uint64_t hca_core_clock; /* in KHZ */
364 struct ib_rss_caps rss_caps;
366 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
367 struct ib_tm_caps tm_caps;
368 struct ib_cq_caps cq_caps;
379 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
382 case IB_MTU_256: return 256;
383 case IB_MTU_512: return 512;
384 case IB_MTU_1024: return 1024;
385 case IB_MTU_2048: return 2048;
386 case IB_MTU_4096: return 4096;
391 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
395 else if (mtu >= 2048)
397 else if (mtu >= 1024)
411 IB_PORT_ACTIVE_DEFER = 5
414 enum ib_port_cap_flags {
416 IB_PORT_NOTICE_SUP = 1 << 2,
417 IB_PORT_TRAP_SUP = 1 << 3,
418 IB_PORT_OPT_IPD_SUP = 1 << 4,
419 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
420 IB_PORT_SL_MAP_SUP = 1 << 6,
421 IB_PORT_MKEY_NVRAM = 1 << 7,
422 IB_PORT_PKEY_NVRAM = 1 << 8,
423 IB_PORT_LED_INFO_SUP = 1 << 9,
424 IB_PORT_SM_DISABLED = 1 << 10,
425 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
426 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
427 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
428 IB_PORT_CM_SUP = 1 << 16,
429 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
430 IB_PORT_REINIT_SUP = 1 << 18,
431 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
432 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
433 IB_PORT_DR_NOTICE_SUP = 1 << 21,
434 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
435 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
436 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
437 IB_PORT_CLIENT_REG_SUP = 1 << 25,
438 IB_PORT_IP_BASED_GIDS = 1 << 26,
448 static inline int ib_width_enum_to_int(enum ib_port_width width)
451 case IB_WIDTH_1X: return 1;
452 case IB_WIDTH_4X: return 4;
453 case IB_WIDTH_8X: return 8;
454 case IB_WIDTH_12X: return 12;
470 * struct rdma_hw_stats
471 * @timestamp - Used by the core code to track when the last update was
472 * @lifespan - Used by the core code to determine how old the counters
473 * should be before being updated again. Stored in jiffies, defaults
474 * to 10 milliseconds, drivers can override the default be specifying
475 * their own value during their allocation routine.
476 * @name - Array of pointers to static names used for the counters in
478 * @num_counters - How many hardware counters there are. If name is
479 * shorter than this number, a kernel oops will result. Driver authors
480 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
481 * in their code to prevent this.
482 * @value - Array of u64 counters that are accessed by the sysfs code and
483 * filled in by the drivers get_stats routine
485 struct rdma_hw_stats {
486 unsigned long timestamp;
487 unsigned long lifespan;
488 const char * const *names;
493 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
495 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
497 * @names - Array of static const char *
498 * @num_counters - How many elements in array
499 * @lifespan - How many milliseconds between updates
501 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
502 const char * const *names, int num_counters,
503 unsigned long lifespan)
505 struct rdma_hw_stats *stats;
507 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
511 stats->names = names;
512 stats->num_counters = num_counters;
513 stats->lifespan = msecs_to_jiffies(lifespan);
519 /* Define bits for the various functionality this port needs to be supported by
522 /* Management 0x00000FFF */
523 #define RDMA_CORE_CAP_IB_MAD 0x00000001
524 #define RDMA_CORE_CAP_IB_SMI 0x00000002
525 #define RDMA_CORE_CAP_IB_CM 0x00000004
526 #define RDMA_CORE_CAP_IW_CM 0x00000008
527 #define RDMA_CORE_CAP_IB_SA 0x00000010
528 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
530 /* Address format 0x000FF000 */
531 #define RDMA_CORE_CAP_AF_IB 0x00001000
532 #define RDMA_CORE_CAP_ETH_AH 0x00002000
533 #define RDMA_CORE_CAP_OPA_AH 0x00004000
535 /* Protocol 0xFFF00000 */
536 #define RDMA_CORE_CAP_PROT_IB 0x00100000
537 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
538 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
539 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
540 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
541 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
543 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
544 | RDMA_CORE_CAP_IB_MAD \
545 | RDMA_CORE_CAP_IB_SMI \
546 | RDMA_CORE_CAP_IB_CM \
547 | RDMA_CORE_CAP_IB_SA \
548 | RDMA_CORE_CAP_AF_IB)
549 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
550 | RDMA_CORE_CAP_IB_MAD \
551 | RDMA_CORE_CAP_IB_CM \
552 | RDMA_CORE_CAP_AF_IB \
553 | RDMA_CORE_CAP_ETH_AH)
554 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
555 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
556 | RDMA_CORE_CAP_IB_MAD \
557 | RDMA_CORE_CAP_IB_CM \
558 | RDMA_CORE_CAP_AF_IB \
559 | RDMA_CORE_CAP_ETH_AH)
560 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
561 | RDMA_CORE_CAP_IW_CM)
562 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
563 | RDMA_CORE_CAP_OPA_MAD)
565 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
567 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
569 struct ib_port_attr {
571 enum ib_port_state state;
573 enum ib_mtu active_mtu;
593 enum ib_device_modify_flags {
594 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
595 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
598 #define IB_DEVICE_NODE_DESC_MAX 64
600 struct ib_device_modify {
602 char node_desc[IB_DEVICE_NODE_DESC_MAX];
605 enum ib_port_modify_flags {
606 IB_PORT_SHUTDOWN = 1,
607 IB_PORT_INIT_TYPE = (1<<2),
608 IB_PORT_RESET_QKEY_CNTR = (1<<3),
609 IB_PORT_OPA_MASK_CHG = (1<<4)
612 struct ib_port_modify {
613 u32 set_port_cap_mask;
614 u32 clr_port_cap_mask;
622 IB_EVENT_QP_ACCESS_ERR,
626 IB_EVENT_PATH_MIG_ERR,
627 IB_EVENT_DEVICE_FATAL,
628 IB_EVENT_PORT_ACTIVE,
631 IB_EVENT_PKEY_CHANGE,
634 IB_EVENT_SRQ_LIMIT_REACHED,
635 IB_EVENT_QP_LAST_WQE_REACHED,
636 IB_EVENT_CLIENT_REREGISTER,
641 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
644 struct ib_device *device;
652 enum ib_event_type event;
655 struct ib_event_handler {
656 struct ib_device *device;
657 void (*handler)(struct ib_event_handler *, struct ib_event *);
658 struct list_head list;
661 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
663 (_ptr)->device = _device; \
664 (_ptr)->handler = _handler; \
665 INIT_LIST_HEAD(&(_ptr)->list); \
668 struct ib_global_route {
677 __be32 version_tclass_flow;
685 union rdma_network_hdr {
688 /* The IB spec states that if it's IPv4, the header
689 * is located in the last 20 bytes of the header.
692 struct iphdr roce4grh;
696 #define IB_QPN_MASK 0xFFFFFF
699 IB_MULTICAST_QPN = 0xffffff
702 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
703 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
710 IB_RATE_PORT_CURRENT = 0,
711 IB_RATE_2_5_GBPS = 2,
719 IB_RATE_120_GBPS = 10,
720 IB_RATE_14_GBPS = 11,
721 IB_RATE_56_GBPS = 12,
722 IB_RATE_112_GBPS = 13,
723 IB_RATE_168_GBPS = 14,
724 IB_RATE_25_GBPS = 15,
725 IB_RATE_100_GBPS = 16,
726 IB_RATE_200_GBPS = 17,
727 IB_RATE_300_GBPS = 18
731 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
732 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
733 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
734 * @rate: rate to convert.
736 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
739 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
740 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
741 * @rate: rate to convert.
743 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
747 * enum ib_mr_type - memory region type
748 * @IB_MR_TYPE_MEM_REG: memory region that is used for
749 * normal registration
750 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
751 * signature operations (data-integrity
753 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
754 * register any arbitrary sg lists (without
755 * the normal mr constraints - see
760 IB_MR_TYPE_SIGNATURE,
766 * IB_SIG_TYPE_NONE: Unprotected.
767 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
769 enum ib_signature_type {
775 * Signature T10-DIF block-guard types
776 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
777 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
779 enum ib_t10_dif_bg_type {
785 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
787 * @bg_type: T10-DIF block guard type (CRC|CSUM)
788 * @pi_interval: protection information interval.
789 * @bg: seed of guard computation.
790 * @app_tag: application tag of guard block
791 * @ref_tag: initial guard block reference tag.
792 * @ref_remap: Indicate wethear the reftag increments each block
793 * @app_escape: Indicate to skip block check if apptag=0xffff
794 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
795 * @apptag_check_mask: check bitmask of application tag.
797 struct ib_t10_dif_domain {
798 enum ib_t10_dif_bg_type bg_type;
806 u16 apptag_check_mask;
810 * struct ib_sig_domain - Parameters for signature domain
811 * @sig_type: specific signauture type
812 * @sig: union of all signature domain attributes that may
813 * be used to set domain layout.
815 struct ib_sig_domain {
816 enum ib_signature_type sig_type;
818 struct ib_t10_dif_domain dif;
823 * struct ib_sig_attrs - Parameters for signature handover operation
824 * @check_mask: bitmask for signature byte check (8 bytes)
825 * @mem: memory domain layout desciptor.
826 * @wire: wire domain layout desciptor.
828 struct ib_sig_attrs {
830 struct ib_sig_domain mem;
831 struct ib_sig_domain wire;
834 enum ib_sig_err_type {
841 * struct ib_sig_err - signature error descriptor
844 enum ib_sig_err_type err_type;
851 enum ib_mr_status_check {
852 IB_MR_CHECK_SIG_STATUS = 1,
856 * struct ib_mr_status - Memory region status container
858 * @fail_status: Bitmask of MR checks status. For each
859 * failed check a corresponding status bit is set.
860 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
863 struct ib_mr_status {
865 struct ib_sig_err sig_err;
869 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
871 * @mult: multiple to convert.
873 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
875 enum rdma_ah_attr_type {
876 RDMA_AH_ATTR_TYPE_IB,
877 RDMA_AH_ATTR_TYPE_ROCE,
878 RDMA_AH_ATTR_TYPE_OPA,
886 struct roce_ah_attr {
896 struct rdma_ah_attr {
897 struct ib_global_route grh;
902 enum rdma_ah_attr_type type;
904 struct ib_ah_attr ib;
905 struct roce_ah_attr roce;
906 struct opa_ah_attr opa;
914 IB_WC_LOC_EEC_OP_ERR,
919 IB_WC_LOC_ACCESS_ERR,
920 IB_WC_REM_INV_REQ_ERR,
921 IB_WC_REM_ACCESS_ERR,
924 IB_WC_RNR_RETRY_EXC_ERR,
925 IB_WC_LOC_RDD_VIOL_ERR,
926 IB_WC_REM_INV_RD_REQ_ERR,
929 IB_WC_INV_EEC_STATE_ERR,
931 IB_WC_RESP_TIMEOUT_ERR,
935 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
946 IB_WC_MASKED_COMP_SWAP,
947 IB_WC_MASKED_FETCH_ADD,
949 * Set value of IB_WC_RECV so consumers can test if a completion is a
950 * receive by testing (opcode & IB_WC_RECV).
953 IB_WC_RECV_RDMA_WITH_IMM
958 IB_WC_WITH_IMM = (1<<1),
959 IB_WC_WITH_INVALIDATE = (1<<2),
960 IB_WC_IP_CSUM_OK = (1<<3),
961 IB_WC_WITH_SMAC = (1<<4),
962 IB_WC_WITH_VLAN = (1<<5),
963 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
969 struct ib_cqe *wr_cqe;
971 enum ib_wc_status status;
972 enum ib_wc_opcode opcode;
986 u8 port_num; /* valid only for DR SMPs on switches */
992 enum ib_cq_notify_flags {
993 IB_CQ_SOLICITED = 1 << 0,
994 IB_CQ_NEXT_COMP = 1 << 1,
995 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
996 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1005 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1007 return srq_type == IB_SRQT_XRC ||
1008 srq_type == IB_SRQT_TM;
1011 enum ib_srq_attr_mask {
1012 IB_SRQ_MAX_WR = 1 << 0,
1013 IB_SRQ_LIMIT = 1 << 1,
1016 struct ib_srq_attr {
1022 struct ib_srq_init_attr {
1023 void (*event_handler)(struct ib_event *, void *);
1025 struct ib_srq_attr attr;
1026 enum ib_srq_type srq_type;
1032 struct ib_xrcd *xrcd;
1047 u32 max_inline_data;
1050 * Maximum number of rdma_rw_ctx structures in flight at a time.
1051 * ib_create_qp() will calculate the right amount of neededed WRs
1052 * and MRs based on this.
1064 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1065 * here (and in that order) since the MAD layer uses them as
1066 * indices into a 2-entry table.
1075 IB_QPT_RAW_ETHERTYPE,
1076 IB_QPT_RAW_PACKET = 8,
1080 IB_QPT_DRIVER = 0xFF,
1081 /* Reserve a range for qp types internal to the low level driver.
1082 * These qp types will not be visible at the IB core layer, so the
1083 * IB_QPT_MAX usages should not be affected in the core layer
1085 IB_QPT_RESERVED1 = 0x1000,
1097 enum ib_qp_create_flags {
1098 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1099 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1100 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1101 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1102 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1103 IB_QP_CREATE_NETIF_QP = 1 << 5,
1104 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1105 /* FREE = 1 << 7, */
1106 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1107 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1108 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1109 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1110 /* reserve bits 26-31 for low level drivers' internal use */
1111 IB_QP_CREATE_RESERVED_START = 1 << 26,
1112 IB_QP_CREATE_RESERVED_END = 1 << 31,
1116 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1117 * callback to destroy the passed in QP.
1120 struct ib_qp_init_attr {
1121 void (*event_handler)(struct ib_event *, void *);
1123 struct ib_cq *send_cq;
1124 struct ib_cq *recv_cq;
1126 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1127 struct ib_qp_cap cap;
1128 enum ib_sig_type sq_sig_type;
1129 enum ib_qp_type qp_type;
1130 enum ib_qp_create_flags create_flags;
1133 * Only needed for special QP types, or when using the RW API.
1136 struct ib_rwq_ind_table *rwq_ind_tbl;
1140 struct ib_qp_open_attr {
1141 void (*event_handler)(struct ib_event *, void *);
1144 enum ib_qp_type qp_type;
1147 enum ib_rnr_timeout {
1148 IB_RNR_TIMER_655_36 = 0,
1149 IB_RNR_TIMER_000_01 = 1,
1150 IB_RNR_TIMER_000_02 = 2,
1151 IB_RNR_TIMER_000_03 = 3,
1152 IB_RNR_TIMER_000_04 = 4,
1153 IB_RNR_TIMER_000_06 = 5,
1154 IB_RNR_TIMER_000_08 = 6,
1155 IB_RNR_TIMER_000_12 = 7,
1156 IB_RNR_TIMER_000_16 = 8,
1157 IB_RNR_TIMER_000_24 = 9,
1158 IB_RNR_TIMER_000_32 = 10,
1159 IB_RNR_TIMER_000_48 = 11,
1160 IB_RNR_TIMER_000_64 = 12,
1161 IB_RNR_TIMER_000_96 = 13,
1162 IB_RNR_TIMER_001_28 = 14,
1163 IB_RNR_TIMER_001_92 = 15,
1164 IB_RNR_TIMER_002_56 = 16,
1165 IB_RNR_TIMER_003_84 = 17,
1166 IB_RNR_TIMER_005_12 = 18,
1167 IB_RNR_TIMER_007_68 = 19,
1168 IB_RNR_TIMER_010_24 = 20,
1169 IB_RNR_TIMER_015_36 = 21,
1170 IB_RNR_TIMER_020_48 = 22,
1171 IB_RNR_TIMER_030_72 = 23,
1172 IB_RNR_TIMER_040_96 = 24,
1173 IB_RNR_TIMER_061_44 = 25,
1174 IB_RNR_TIMER_081_92 = 26,
1175 IB_RNR_TIMER_122_88 = 27,
1176 IB_RNR_TIMER_163_84 = 28,
1177 IB_RNR_TIMER_245_76 = 29,
1178 IB_RNR_TIMER_327_68 = 30,
1179 IB_RNR_TIMER_491_52 = 31
1182 enum ib_qp_attr_mask {
1184 IB_QP_CUR_STATE = (1<<1),
1185 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1186 IB_QP_ACCESS_FLAGS = (1<<3),
1187 IB_QP_PKEY_INDEX = (1<<4),
1188 IB_QP_PORT = (1<<5),
1189 IB_QP_QKEY = (1<<6),
1191 IB_QP_PATH_MTU = (1<<8),
1192 IB_QP_TIMEOUT = (1<<9),
1193 IB_QP_RETRY_CNT = (1<<10),
1194 IB_QP_RNR_RETRY = (1<<11),
1195 IB_QP_RQ_PSN = (1<<12),
1196 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1197 IB_QP_ALT_PATH = (1<<14),
1198 IB_QP_MIN_RNR_TIMER = (1<<15),
1199 IB_QP_SQ_PSN = (1<<16),
1200 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1201 IB_QP_PATH_MIG_STATE = (1<<18),
1202 IB_QP_CAP = (1<<19),
1203 IB_QP_DEST_QPN = (1<<20),
1204 IB_QP_RESERVED1 = (1<<21),
1205 IB_QP_RESERVED2 = (1<<22),
1206 IB_QP_RESERVED3 = (1<<23),
1207 IB_QP_RESERVED4 = (1<<24),
1208 IB_QP_RATE_LIMIT = (1<<25),
1233 enum ib_qp_state qp_state;
1234 enum ib_qp_state cur_qp_state;
1235 enum ib_mtu path_mtu;
1236 enum ib_mig_state path_mig_state;
1241 int qp_access_flags;
1242 struct ib_qp_cap cap;
1243 struct rdma_ah_attr ah_attr;
1244 struct rdma_ah_attr alt_ah_attr;
1247 u8 en_sqd_async_notify;
1250 u8 max_dest_rd_atomic;
1263 IB_WR_RDMA_WRITE_WITH_IMM,
1265 IB_WR_SEND_WITH_IMM,
1267 IB_WR_ATOMIC_CMP_AND_SWP,
1268 IB_WR_ATOMIC_FETCH_AND_ADD,
1270 IB_WR_SEND_WITH_INV,
1271 IB_WR_RDMA_READ_WITH_INV,
1274 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1275 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1277 /* reserve values for low level drivers' internal use.
1278 * These values will not be used at all in the ib core layer.
1280 IB_WR_RESERVED1 = 0xf0,
1292 enum ib_send_flags {
1294 IB_SEND_SIGNALED = (1<<1),
1295 IB_SEND_SOLICITED = (1<<2),
1296 IB_SEND_INLINE = (1<<3),
1297 IB_SEND_IP_CSUM = (1<<4),
1299 /* reserve bits 26-31 for low level drivers' internal use */
1300 IB_SEND_RESERVED_START = (1 << 26),
1301 IB_SEND_RESERVED_END = (1 << 31),
1311 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1315 struct ib_send_wr *next;
1318 struct ib_cqe *wr_cqe;
1320 struct ib_sge *sg_list;
1322 enum ib_wr_opcode opcode;
1326 u32 invalidate_rkey;
1331 struct ib_send_wr wr;
1336 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1338 return container_of(wr, struct ib_rdma_wr, wr);
1341 struct ib_atomic_wr {
1342 struct ib_send_wr wr;
1346 u64 compare_add_mask;
1351 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1353 return container_of(wr, struct ib_atomic_wr, wr);
1357 struct ib_send_wr wr;
1364 u16 pkey_index; /* valid for GSI only */
1365 u8 port_num; /* valid for DR SMPs on switch only */
1368 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1370 return container_of(wr, struct ib_ud_wr, wr);
1374 struct ib_send_wr wr;
1380 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1382 return container_of(wr, struct ib_reg_wr, wr);
1385 struct ib_sig_handover_wr {
1386 struct ib_send_wr wr;
1387 struct ib_sig_attrs *sig_attrs;
1388 struct ib_mr *sig_mr;
1390 struct ib_sge *prot;
1393 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1395 return container_of(wr, struct ib_sig_handover_wr, wr);
1399 struct ib_recv_wr *next;
1402 struct ib_cqe *wr_cqe;
1404 struct ib_sge *sg_list;
1408 enum ib_access_flags {
1409 IB_ACCESS_LOCAL_WRITE = 1,
1410 IB_ACCESS_REMOTE_WRITE = (1<<1),
1411 IB_ACCESS_REMOTE_READ = (1<<2),
1412 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1413 IB_ACCESS_MW_BIND = (1<<4),
1414 IB_ZERO_BASED = (1<<5),
1415 IB_ACCESS_ON_DEMAND = (1<<6),
1416 IB_ACCESS_HUGETLB = (1<<7),
1420 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1421 * are hidden here instead of a uapi header!
1423 enum ib_mr_rereg_flags {
1424 IB_MR_REREG_TRANS = 1,
1425 IB_MR_REREG_PD = (1<<1),
1426 IB_MR_REREG_ACCESS = (1<<2),
1427 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1430 struct ib_fmr_attr {
1438 enum rdma_remove_reason {
1439 /* Userspace requested uobject deletion. Call could fail */
1440 RDMA_REMOVE_DESTROY,
1441 /* Context deletion. This call should delete the actual object itself */
1443 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1444 RDMA_REMOVE_DRIVER_REMOVE,
1445 /* Context is being cleaned-up, but commit was just completed */
1446 RDMA_REMOVE_DURING_CLEANUP,
1449 struct ib_rdmacg_object {
1450 #ifdef CONFIG_CGROUP_RDMA
1451 struct rdma_cgroup *cg; /* owner rdma cgroup */
1455 struct ib_ucontext {
1456 struct ib_device *device;
1457 struct ib_uverbs_file *ufile;
1460 /* locking the uobjects_list */
1461 struct mutex uobjects_lock;
1462 struct list_head uobjects;
1463 /* protects cleanup process from other actions */
1464 struct rw_semaphore cleanup_rwsem;
1465 enum rdma_remove_reason cleanup_reason;
1468 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1469 struct rb_root_cached umem_tree;
1471 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1472 * mmu notifiers registration.
1474 struct rw_semaphore umem_rwsem;
1475 void (*invalidate_range)(struct ib_umem *umem,
1476 unsigned long start, unsigned long end);
1478 struct mmu_notifier mn;
1479 atomic_t notifier_count;
1480 /* A list of umems that don't have private mmu notifier counters yet. */
1481 struct list_head no_private_counters;
1485 struct ib_rdmacg_object cg_obj;
1489 u64 user_handle; /* handle given to us by userspace */
1490 struct ib_ucontext *context; /* associated user context */
1491 void *object; /* containing object */
1492 struct list_head list; /* link to context's list */
1493 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1494 int id; /* index into kernel idr */
1496 atomic_t usecnt; /* protects exclusive access */
1497 struct rcu_head rcu; /* kfree_rcu() overhead */
1499 const struct uverbs_obj_type *type;
1502 struct ib_uobject_file {
1503 struct ib_uobject uobj;
1504 /* ufile contains the lock between context release and file close */
1505 struct ib_uverbs_file *ufile;
1509 const void __user *inbuf;
1510 void __user *outbuf;
1518 struct ib_device *device;
1519 struct ib_uobject *uobject;
1520 atomic_t usecnt; /* count all resources */
1522 u32 unsafe_global_rkey;
1525 * Implementation details of the RDMA core, don't use in drivers:
1527 struct ib_mr *__internal_mr;
1531 struct ib_device *device;
1532 atomic_t usecnt; /* count all exposed resources */
1533 struct inode *inode;
1535 struct mutex tgt_qp_mutex;
1536 struct list_head tgt_qp_list;
1540 struct ib_device *device;
1542 struct ib_uobject *uobject;
1543 enum rdma_ah_attr_type type;
1546 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1548 enum ib_poll_context {
1549 IB_POLL_DIRECT, /* caller context, no hw completions */
1550 IB_POLL_SOFTIRQ, /* poll from softirq context */
1551 IB_POLL_WORKQUEUE, /* poll from workqueue */
1555 struct ib_device *device;
1556 struct ib_uobject *uobject;
1557 ib_comp_handler comp_handler;
1558 void (*event_handler)(struct ib_event *, void *);
1561 atomic_t usecnt; /* count number of work queues */
1562 enum ib_poll_context poll_ctx;
1565 struct irq_poll iop;
1566 struct work_struct work;
1571 struct ib_device *device;
1573 struct ib_uobject *uobject;
1574 void (*event_handler)(struct ib_event *, void *);
1576 enum ib_srq_type srq_type;
1583 struct ib_xrcd *xrcd;
1590 enum ib_raw_packet_caps {
1591 /* Strip cvlan from incoming packet and report it in the matching work
1592 * completion is supported.
1594 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1595 /* Scatter FCS field of an incoming packet to host memory is supported.
1597 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1598 /* Checksum offloads are supported (for both send and receive). */
1599 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1600 /* When a packet is received for an RQ with no receive WQEs, the
1601 * packet processing is delayed.
1603 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1617 struct ib_device *device;
1618 struct ib_uobject *uobject;
1620 void (*event_handler)(struct ib_event *, void *);
1624 enum ib_wq_state state;
1625 enum ib_wq_type wq_type;
1630 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1631 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1632 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1633 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1636 struct ib_wq_init_attr {
1638 enum ib_wq_type wq_type;
1642 void (*event_handler)(struct ib_event *, void *);
1643 u32 create_flags; /* Use enum ib_wq_flags */
1646 enum ib_wq_attr_mask {
1647 IB_WQ_STATE = 1 << 0,
1648 IB_WQ_CUR_STATE = 1 << 1,
1649 IB_WQ_FLAGS = 1 << 2,
1653 enum ib_wq_state wq_state;
1654 enum ib_wq_state curr_wq_state;
1655 u32 flags; /* Use enum ib_wq_flags */
1656 u32 flags_mask; /* Use enum ib_wq_flags */
1659 struct ib_rwq_ind_table {
1660 struct ib_device *device;
1661 struct ib_uobject *uobject;
1664 u32 log_ind_tbl_size;
1665 struct ib_wq **ind_tbl;
1668 struct ib_rwq_ind_table_init_attr {
1669 u32 log_ind_tbl_size;
1670 /* Each entry is a pointer to Receive Work Queue */
1671 struct ib_wq **ind_tbl;
1674 enum port_pkey_state {
1675 IB_PORT_PKEY_NOT_VALID = 0,
1676 IB_PORT_PKEY_VALID = 1,
1677 IB_PORT_PKEY_LISTED = 2,
1680 struct ib_qp_security;
1682 struct ib_port_pkey {
1683 enum port_pkey_state state;
1686 struct list_head qp_list;
1687 struct list_head to_error_list;
1688 struct ib_qp_security *sec;
1691 struct ib_ports_pkeys {
1692 struct ib_port_pkey main;
1693 struct ib_port_pkey alt;
1696 struct ib_qp_security {
1698 struct ib_device *dev;
1699 /* Hold this mutex when changing port and pkey settings. */
1701 struct ib_ports_pkeys *ports_pkeys;
1702 /* A list of all open shared QP handles. Required to enforce security
1703 * properly for all users of a shared QP.
1705 struct list_head shared_qp_list;
1708 atomic_t error_list_count;
1709 struct completion error_complete;
1710 int error_comps_pending;
1714 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1715 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1718 struct ib_device *device;
1720 struct ib_cq *send_cq;
1721 struct ib_cq *recv_cq;
1724 struct list_head rdma_mrs;
1725 struct list_head sig_mrs;
1727 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1728 struct list_head xrcd_list;
1730 /* count times opened, mcast attaches, flow attaches */
1732 struct list_head open_list;
1733 struct ib_qp *real_qp;
1734 struct ib_uobject *uobject;
1735 void (*event_handler)(struct ib_event *, void *);
1740 enum ib_qp_type qp_type;
1741 struct ib_rwq_ind_table *rwq_ind_tbl;
1742 struct ib_qp_security *qp_sec;
1747 struct ib_device *device;
1753 unsigned int page_size;
1756 struct ib_uobject *uobject; /* user */
1757 struct list_head qp_entry; /* FR */
1762 struct ib_device *device;
1764 struct ib_uobject *uobject;
1766 enum ib_mw_type type;
1770 struct ib_device *device;
1772 struct list_head list;
1777 /* Supported steering options */
1778 enum ib_flow_attr_type {
1779 /* steering according to rule specifications */
1780 IB_FLOW_ATTR_NORMAL = 0x0,
1781 /* default unicast and multicast rule -
1782 * receive all Eth traffic which isn't steered to any QP
1784 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1785 /* default multicast rule -
1786 * receive all Eth multicast traffic which isn't steered to any QP
1788 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1789 /* sniffer rule - receive all port traffic */
1790 IB_FLOW_ATTR_SNIFFER = 0x3
1793 /* Supported steering header types */
1794 enum ib_flow_spec_type {
1796 IB_FLOW_SPEC_ETH = 0x20,
1797 IB_FLOW_SPEC_IB = 0x22,
1799 IB_FLOW_SPEC_IPV4 = 0x30,
1800 IB_FLOW_SPEC_IPV6 = 0x31,
1802 IB_FLOW_SPEC_TCP = 0x40,
1803 IB_FLOW_SPEC_UDP = 0x41,
1804 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1805 IB_FLOW_SPEC_INNER = 0x100,
1807 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1808 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1810 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1811 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1813 /* Flow steering rule priority is set according to it's domain.
1814 * Lower domain value means higher priority.
1816 enum ib_flow_domain {
1817 IB_FLOW_DOMAIN_USER,
1818 IB_FLOW_DOMAIN_ETHTOOL,
1821 IB_FLOW_DOMAIN_NUM /* Must be last */
1824 enum ib_flow_flags {
1825 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1826 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1829 struct ib_flow_eth_filter {
1838 struct ib_flow_spec_eth {
1841 struct ib_flow_eth_filter val;
1842 struct ib_flow_eth_filter mask;
1845 struct ib_flow_ib_filter {
1852 struct ib_flow_spec_ib {
1855 struct ib_flow_ib_filter val;
1856 struct ib_flow_ib_filter mask;
1859 /* IPv4 header flags */
1860 enum ib_ipv4_flags {
1861 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1862 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1863 last have this flag set */
1866 struct ib_flow_ipv4_filter {
1877 struct ib_flow_spec_ipv4 {
1880 struct ib_flow_ipv4_filter val;
1881 struct ib_flow_ipv4_filter mask;
1884 struct ib_flow_ipv6_filter {
1895 struct ib_flow_spec_ipv6 {
1898 struct ib_flow_ipv6_filter val;
1899 struct ib_flow_ipv6_filter mask;
1902 struct ib_flow_tcp_udp_filter {
1909 struct ib_flow_spec_tcp_udp {
1912 struct ib_flow_tcp_udp_filter val;
1913 struct ib_flow_tcp_udp_filter mask;
1916 struct ib_flow_tunnel_filter {
1921 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1922 * the tunnel_id from val has the vni value
1924 struct ib_flow_spec_tunnel {
1927 struct ib_flow_tunnel_filter val;
1928 struct ib_flow_tunnel_filter mask;
1931 struct ib_flow_spec_action_tag {
1932 enum ib_flow_spec_type type;
1937 struct ib_flow_spec_action_drop {
1938 enum ib_flow_spec_type type;
1942 union ib_flow_spec {
1947 struct ib_flow_spec_eth eth;
1948 struct ib_flow_spec_ib ib;
1949 struct ib_flow_spec_ipv4 ipv4;
1950 struct ib_flow_spec_tcp_udp tcp_udp;
1951 struct ib_flow_spec_ipv6 ipv6;
1952 struct ib_flow_spec_tunnel tunnel;
1953 struct ib_flow_spec_action_tag flow_tag;
1954 struct ib_flow_spec_action_drop drop;
1957 struct ib_flow_attr {
1958 enum ib_flow_attr_type type;
1964 /* Following are the optional layers according to user request
1965 * struct ib_flow_spec_xxx
1966 * struct ib_flow_spec_yyy
1972 struct ib_uobject *uobject;
1978 enum ib_process_mad_flags {
1979 IB_MAD_IGNORE_MKEY = 1,
1980 IB_MAD_IGNORE_BKEY = 2,
1981 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1984 enum ib_mad_result {
1985 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1986 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1987 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1988 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1991 struct ib_port_cache {
1993 struct ib_pkey_cache *pkey;
1994 struct ib_gid_table *gid;
1996 enum ib_port_state port_state;
2001 struct ib_event_handler event_handler;
2002 struct ib_port_cache *ports;
2007 struct ib_port_immutable {
2014 /* rdma netdev type - specifies protocol type */
2015 enum rdma_netdev_t {
2016 RDMA_NETDEV_OPA_VNIC,
2021 * struct rdma_netdev - rdma netdev
2022 * For cases where netstack interfacing is required.
2024 struct rdma_netdev {
2026 struct ib_device *hca;
2029 /* cleanup function must be specified */
2030 void (*free_rdma_netdev)(struct net_device *netdev);
2032 /* control functions */
2033 void (*set_id)(struct net_device *netdev, int id);
2035 int (*send)(struct net_device *dev, struct sk_buff *skb,
2036 struct ib_ah *address, u32 dqpn);
2038 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2039 union ib_gid *gid, u16 mlid,
2040 int set_qkey, u32 qkey);
2041 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2042 union ib_gid *gid, u16 mlid);
2045 struct ib_port_pkey_list {
2046 /* Lock to hold while modifying the list. */
2047 spinlock_t list_lock;
2048 struct list_head pkey_list;
2052 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2053 struct device *dma_device;
2055 char name[IB_DEVICE_NAME_MAX];
2057 struct list_head event_handler_list;
2058 spinlock_t event_handler_lock;
2060 spinlock_t client_data_lock;
2061 struct list_head core_list;
2062 /* Access to the client_data_list is protected by the client_data_lock
2063 * spinlock and the lists_rwsem read-write semaphore */
2064 struct list_head client_data_list;
2066 struct ib_cache cache;
2068 * port_immutable is indexed by port number
2070 struct ib_port_immutable *port_immutable;
2072 int num_comp_vectors;
2074 struct ib_port_pkey_list *port_pkey_list;
2076 struct iw_cm_verbs *iwcm;
2079 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2080 * driver initialized data. The struct is kfree()'ed by the sysfs
2081 * core when the device is removed. A lifespan of -1 in the return
2082 * struct tells the core to set a default lifespan.
2084 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2087 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2088 * @index - The index in the value array we wish to have updated, or
2089 * num_counters if we want all stats updated
2091 * < 0 - Error, no counters updated
2092 * index - Updated the single counter pointed to by index
2093 * num_counters - Updated all counters (will reset the timestamp
2094 * and prevent further calls for lifespan milliseconds)
2095 * Drivers are allowed to update all counters in leiu of just the
2096 * one given in index at their option
2098 int (*get_hw_stats)(struct ib_device *device,
2099 struct rdma_hw_stats *stats,
2100 u8 port, int index);
2101 int (*query_device)(struct ib_device *device,
2102 struct ib_device_attr *device_attr,
2103 struct ib_udata *udata);
2104 int (*query_port)(struct ib_device *device,
2106 struct ib_port_attr *port_attr);
2107 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2109 /* When calling get_netdev, the HW vendor's driver should return the
2110 * net device of device @device at port @port_num or NULL if such
2111 * a net device doesn't exist. The vendor driver should call dev_hold
2112 * on this net device. The HW vendor's device driver must guarantee
2113 * that this function returns NULL before the net device reaches
2114 * NETDEV_UNREGISTER_FINAL state.
2116 struct net_device *(*get_netdev)(struct ib_device *device,
2118 int (*query_gid)(struct ib_device *device,
2119 u8 port_num, int index,
2121 /* When calling add_gid, the HW vendor's driver should
2122 * add the gid of device @device at gid index @index of
2123 * port @port_num to be @gid. Meta-info of that gid (for example,
2124 * the network device related to this gid is available
2125 * at @attr. @context allows the HW vendor driver to store extra
2126 * information together with a GID entry. The HW vendor may allocate
2127 * memory to contain this information and store it in @context when a
2128 * new GID entry is written to. Params are consistent until the next
2129 * call of add_gid or delete_gid. The function should return 0 on
2130 * success or error otherwise. The function could be called
2131 * concurrently for different ports. This function is only called
2132 * when roce_gid_table is used.
2134 int (*add_gid)(struct ib_device *device,
2137 const union ib_gid *gid,
2138 const struct ib_gid_attr *attr,
2140 /* When calling del_gid, the HW vendor's driver should delete the
2141 * gid of device @device at gid index @index of port @port_num.
2142 * Upon the deletion of a GID entry, the HW vendor must free any
2143 * allocated memory. The caller will clear @context afterwards.
2144 * This function is only called when roce_gid_table is used.
2146 int (*del_gid)(struct ib_device *device,
2150 int (*query_pkey)(struct ib_device *device,
2151 u8 port_num, u16 index, u16 *pkey);
2152 int (*modify_device)(struct ib_device *device,
2153 int device_modify_mask,
2154 struct ib_device_modify *device_modify);
2155 int (*modify_port)(struct ib_device *device,
2156 u8 port_num, int port_modify_mask,
2157 struct ib_port_modify *port_modify);
2158 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2159 struct ib_udata *udata);
2160 int (*dealloc_ucontext)(struct ib_ucontext *context);
2161 int (*mmap)(struct ib_ucontext *context,
2162 struct vm_area_struct *vma);
2163 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2164 struct ib_ucontext *context,
2165 struct ib_udata *udata);
2166 int (*dealloc_pd)(struct ib_pd *pd);
2167 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2168 struct rdma_ah_attr *ah_attr,
2169 struct ib_udata *udata);
2170 int (*modify_ah)(struct ib_ah *ah,
2171 struct rdma_ah_attr *ah_attr);
2172 int (*query_ah)(struct ib_ah *ah,
2173 struct rdma_ah_attr *ah_attr);
2174 int (*destroy_ah)(struct ib_ah *ah);
2175 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2176 struct ib_srq_init_attr *srq_init_attr,
2177 struct ib_udata *udata);
2178 int (*modify_srq)(struct ib_srq *srq,
2179 struct ib_srq_attr *srq_attr,
2180 enum ib_srq_attr_mask srq_attr_mask,
2181 struct ib_udata *udata);
2182 int (*query_srq)(struct ib_srq *srq,
2183 struct ib_srq_attr *srq_attr);
2184 int (*destroy_srq)(struct ib_srq *srq);
2185 int (*post_srq_recv)(struct ib_srq *srq,
2186 struct ib_recv_wr *recv_wr,
2187 struct ib_recv_wr **bad_recv_wr);
2188 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2189 struct ib_qp_init_attr *qp_init_attr,
2190 struct ib_udata *udata);
2191 int (*modify_qp)(struct ib_qp *qp,
2192 struct ib_qp_attr *qp_attr,
2194 struct ib_udata *udata);
2195 int (*query_qp)(struct ib_qp *qp,
2196 struct ib_qp_attr *qp_attr,
2198 struct ib_qp_init_attr *qp_init_attr);
2199 int (*destroy_qp)(struct ib_qp *qp);
2200 int (*post_send)(struct ib_qp *qp,
2201 struct ib_send_wr *send_wr,
2202 struct ib_send_wr **bad_send_wr);
2203 int (*post_recv)(struct ib_qp *qp,
2204 struct ib_recv_wr *recv_wr,
2205 struct ib_recv_wr **bad_recv_wr);
2206 struct ib_cq * (*create_cq)(struct ib_device *device,
2207 const struct ib_cq_init_attr *attr,
2208 struct ib_ucontext *context,
2209 struct ib_udata *udata);
2210 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2212 int (*destroy_cq)(struct ib_cq *cq);
2213 int (*resize_cq)(struct ib_cq *cq, int cqe,
2214 struct ib_udata *udata);
2215 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2217 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2218 int (*req_notify_cq)(struct ib_cq *cq,
2219 enum ib_cq_notify_flags flags);
2220 int (*req_ncomp_notif)(struct ib_cq *cq,
2222 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2223 int mr_access_flags);
2224 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2225 u64 start, u64 length,
2227 int mr_access_flags,
2228 struct ib_udata *udata);
2229 int (*rereg_user_mr)(struct ib_mr *mr,
2231 u64 start, u64 length,
2233 int mr_access_flags,
2235 struct ib_udata *udata);
2236 int (*dereg_mr)(struct ib_mr *mr);
2237 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2238 enum ib_mr_type mr_type,
2240 int (*map_mr_sg)(struct ib_mr *mr,
2241 struct scatterlist *sg,
2243 unsigned int *sg_offset);
2244 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2245 enum ib_mw_type type,
2246 struct ib_udata *udata);
2247 int (*dealloc_mw)(struct ib_mw *mw);
2248 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2249 int mr_access_flags,
2250 struct ib_fmr_attr *fmr_attr);
2251 int (*map_phys_fmr)(struct ib_fmr *fmr,
2252 u64 *page_list, int list_len,
2254 int (*unmap_fmr)(struct list_head *fmr_list);
2255 int (*dealloc_fmr)(struct ib_fmr *fmr);
2256 int (*attach_mcast)(struct ib_qp *qp,
2259 int (*detach_mcast)(struct ib_qp *qp,
2262 int (*process_mad)(struct ib_device *device,
2263 int process_mad_flags,
2265 const struct ib_wc *in_wc,
2266 const struct ib_grh *in_grh,
2267 const struct ib_mad_hdr *in_mad,
2269 struct ib_mad_hdr *out_mad,
2270 size_t *out_mad_size,
2271 u16 *out_mad_pkey_index);
2272 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2273 struct ib_ucontext *ucontext,
2274 struct ib_udata *udata);
2275 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2276 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2280 int (*destroy_flow)(struct ib_flow *flow_id);
2281 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2282 struct ib_mr_status *mr_status);
2283 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2284 void (*drain_rq)(struct ib_qp *qp);
2285 void (*drain_sq)(struct ib_qp *qp);
2286 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2288 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2289 struct ifla_vf_info *ivf);
2290 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2291 struct ifla_vf_stats *stats);
2292 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2294 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2295 struct ib_wq_init_attr *init_attr,
2296 struct ib_udata *udata);
2297 int (*destroy_wq)(struct ib_wq *wq);
2298 int (*modify_wq)(struct ib_wq *wq,
2299 struct ib_wq_attr *attr,
2301 struct ib_udata *udata);
2302 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2303 struct ib_rwq_ind_table_init_attr *init_attr,
2304 struct ib_udata *udata);
2305 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2307 * rdma netdev operation
2309 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2310 * doesn't support the specified rdma netdev type.
2312 struct net_device *(*alloc_rdma_netdev)(
2313 struct ib_device *device,
2315 enum rdma_netdev_t type,
2317 unsigned char name_assign_type,
2318 void (*setup)(struct net_device *));
2320 struct module *owner;
2322 struct kobject *ports_parent;
2323 struct list_head port_list;
2326 IB_DEV_UNINITIALIZED,
2332 u64 uverbs_cmd_mask;
2333 u64 uverbs_ex_cmd_mask;
2335 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2341 struct ib_device_attr attrs;
2342 struct attribute_group *hw_stats_ag;
2343 struct rdma_hw_stats *hw_stats;
2345 #ifdef CONFIG_CGROUP_RDMA
2346 struct rdmacg_device cg_device;
2352 * The following mandatory functions are used only at device
2353 * registration. Keep functions such as these at the end of this
2354 * structure to avoid cache line misses when accessing struct ib_device
2357 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2358 void (*get_dev_fw_str)(struct ib_device *, char *str);
2359 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2362 struct uverbs_root_spec *specs_root;
2367 void (*add) (struct ib_device *);
2368 void (*remove)(struct ib_device *, void *client_data);
2370 /* Returns the net_dev belonging to this ib_client and matching the
2372 * @dev: An RDMA device that the net_dev use for communication.
2373 * @port: A physical port number on the RDMA device.
2374 * @pkey: P_Key that the net_dev uses if applicable.
2375 * @gid: A GID that the net_dev uses to communicate.
2376 * @addr: An IP address the net_dev is configured with.
2377 * @client_data: The device's client data set by ib_set_client_data().
2379 * An ib_client that implements a net_dev on top of RDMA devices
2380 * (such as IP over IB) should implement this callback, allowing the
2381 * rdma_cm module to find the right net_dev for a given request.
2383 * The caller is responsible for calling dev_put on the returned
2385 struct net_device *(*get_net_dev_by_params)(
2386 struct ib_device *dev,
2389 const union ib_gid *gid,
2390 const struct sockaddr *addr,
2392 struct list_head list;
2395 struct ib_device *ib_alloc_device(size_t size);
2396 void ib_dealloc_device(struct ib_device *device);
2398 void ib_get_device_fw_str(struct ib_device *device, char *str);
2400 int ib_register_device(struct ib_device *device,
2401 int (*port_callback)(struct ib_device *,
2402 u8, struct kobject *));
2403 void ib_unregister_device(struct ib_device *device);
2405 int ib_register_client (struct ib_client *client);
2406 void ib_unregister_client(struct ib_client *client);
2408 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2409 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2412 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2414 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2417 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2419 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2422 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2426 const void __user *p = udata->inbuf + offset;
2430 if (len > USHRT_MAX)
2433 buf = memdup_user(p, len);
2437 ret = !memchr_inv(buf, 0, len);
2443 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2444 * contains all required attributes and no attributes not allowed for
2445 * the given QP state transition.
2446 * @cur_state: Current QP state
2447 * @next_state: Next QP state
2449 * @mask: Mask of supplied QP attributes
2450 * @ll : link layer of port
2452 * This function is a helper function that a low-level driver's
2453 * modify_qp method can use to validate the consumer's input. It
2454 * checks that cur_state and next_state are valid QP states, that a
2455 * transition from cur_state to next_state is allowed by the IB spec,
2456 * and that the attribute mask supplied is allowed for the transition.
2458 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2459 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2460 enum rdma_link_layer ll);
2462 void ib_register_event_handler(struct ib_event_handler *event_handler);
2463 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2464 void ib_dispatch_event(struct ib_event *event);
2466 int ib_query_port(struct ib_device *device,
2467 u8 port_num, struct ib_port_attr *port_attr);
2469 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2473 * rdma_cap_ib_switch - Check if the device is IB switch
2474 * @device: Device to check
2476 * Device driver is responsible for setting is_switch bit on
2477 * in ib_device structure at init time.
2479 * Return: true if the device is IB switch.
2481 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2483 return device->is_switch;
2487 * rdma_start_port - Return the first valid port number for the device
2490 * @device: Device to be checked
2492 * Return start port number
2494 static inline u8 rdma_start_port(const struct ib_device *device)
2496 return rdma_cap_ib_switch(device) ? 0 : 1;
2500 * rdma_end_port - Return the last valid port number for the device
2503 * @device: Device to be checked
2505 * Return last port number
2507 static inline u8 rdma_end_port(const struct ib_device *device)
2509 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2512 static inline int rdma_is_port_valid(const struct ib_device *device,
2515 return (port >= rdma_start_port(device) &&
2516 port <= rdma_end_port(device));
2519 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2521 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2524 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2526 return device->port_immutable[port_num].core_cap_flags &
2527 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2530 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2532 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2535 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2537 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2540 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2542 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2545 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2547 return rdma_protocol_ib(device, port_num) ||
2548 rdma_protocol_roce(device, port_num);
2551 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2553 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2556 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2558 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2562 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2563 * Management Datagrams.
2564 * @device: Device to check
2565 * @port_num: Port number to check
2567 * Management Datagrams (MAD) are a required part of the InfiniBand
2568 * specification and are supported on all InfiniBand devices. A slightly
2569 * extended version are also supported on OPA interfaces.
2571 * Return: true if the port supports sending/receiving of MAD packets.
2573 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2575 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2579 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2580 * Management Datagrams.
2581 * @device: Device to check
2582 * @port_num: Port number to check
2584 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2585 * datagrams with their own versions. These OPA MADs share many but not all of
2586 * the characteristics of InfiniBand MADs.
2588 * OPA MADs differ in the following ways:
2590 * 1) MADs are variable size up to 2K
2591 * IBTA defined MADs remain fixed at 256 bytes
2592 * 2) OPA SMPs must carry valid PKeys
2593 * 3) OPA SMP packets are a different format
2595 * Return: true if the port supports OPA MAD packet formats.
2597 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2599 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2600 == RDMA_CORE_CAP_OPA_MAD;
2604 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2605 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2606 * @device: Device to check
2607 * @port_num: Port number to check
2609 * Each InfiniBand node is required to provide a Subnet Management Agent
2610 * that the subnet manager can access. Prior to the fabric being fully
2611 * configured by the subnet manager, the SMA is accessed via a well known
2612 * interface called the Subnet Management Interface (SMI). This interface
2613 * uses directed route packets to communicate with the SM to get around the
2614 * chicken and egg problem of the SM needing to know what's on the fabric
2615 * in order to configure the fabric, and needing to configure the fabric in
2616 * order to send packets to the devices on the fabric. These directed
2617 * route packets do not need the fabric fully configured in order to reach
2618 * their destination. The SMI is the only method allowed to send
2619 * directed route packets on an InfiniBand fabric.
2621 * Return: true if the port provides an SMI.
2623 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2625 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2629 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2630 * Communication Manager.
2631 * @device: Device to check
2632 * @port_num: Port number to check
2634 * The InfiniBand Communication Manager is one of many pre-defined General
2635 * Service Agents (GSA) that are accessed via the General Service
2636 * Interface (GSI). It's role is to facilitate establishment of connections
2637 * between nodes as well as other management related tasks for established
2640 * Return: true if the port supports an IB CM (this does not guarantee that
2641 * a CM is actually running however).
2643 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2645 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2649 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2650 * Communication Manager.
2651 * @device: Device to check
2652 * @port_num: Port number to check
2654 * Similar to above, but specific to iWARP connections which have a different
2655 * managment protocol than InfiniBand.
2657 * Return: true if the port supports an iWARP CM (this does not guarantee that
2658 * a CM is actually running however).
2660 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2662 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2666 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2667 * Subnet Administration.
2668 * @device: Device to check
2669 * @port_num: Port number to check
2671 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2672 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2673 * fabrics, devices should resolve routes to other hosts by contacting the
2674 * SA to query the proper route.
2676 * Return: true if the port should act as a client to the fabric Subnet
2677 * Administration interface. This does not imply that the SA service is
2680 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2682 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2686 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2688 * @device: Device to check
2689 * @port_num: Port number to check
2691 * InfiniBand multicast registration is more complex than normal IPv4 or
2692 * IPv6 multicast registration. Each Host Channel Adapter must register
2693 * with the Subnet Manager when it wishes to join a multicast group. It
2694 * should do so only once regardless of how many queue pairs it subscribes
2695 * to this group. And it should leave the group only after all queue pairs
2696 * attached to the group have been detached.
2698 * Return: true if the port must undertake the additional adminstrative
2699 * overhead of registering/unregistering with the SM and tracking of the
2700 * total number of queue pairs attached to the multicast group.
2702 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2704 return rdma_cap_ib_sa(device, port_num);
2708 * rdma_cap_af_ib - Check if the port of device has the capability
2709 * Native Infiniband Address.
2710 * @device: Device to check
2711 * @port_num: Port number to check
2713 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2714 * GID. RoCE uses a different mechanism, but still generates a GID via
2715 * a prescribed mechanism and port specific data.
2717 * Return: true if the port uses a GID address to identify devices on the
2720 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2722 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2726 * rdma_cap_eth_ah - Check if the port of device has the capability
2727 * Ethernet Address Handle.
2728 * @device: Device to check
2729 * @port_num: Port number to check
2731 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2732 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2733 * port. Normally, packet headers are generated by the sending host
2734 * adapter, but when sending connectionless datagrams, we must manually
2735 * inject the proper headers for the fabric we are communicating over.
2737 * Return: true if we are running as a RoCE port and must force the
2738 * addition of a Global Route Header built from our Ethernet Address
2739 * Handle into our header list for connectionless packets.
2741 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2743 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2747 * rdma_cap_opa_ah - Check if the port of device supports
2748 * OPA Address handles
2749 * @device: Device to check
2750 * @port_num: Port number to check
2752 * Return: true if we are running on an OPA device which supports
2753 * the extended OPA addressing.
2755 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2757 return (device->port_immutable[port_num].core_cap_flags &
2758 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2762 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2765 * @port_num: Port number
2767 * This MAD size includes the MAD headers and MAD payload. No other headers
2770 * Return the max MAD size required by the Port. Will return 0 if the port
2771 * does not support MADs
2773 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2775 return device->port_immutable[port_num].max_mad_size;
2779 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2780 * @device: Device to check
2781 * @port_num: Port number to check
2783 * RoCE GID table mechanism manages the various GIDs for a device.
2785 * NOTE: if allocating the port's GID table has failed, this call will still
2786 * return true, but any RoCE GID table API will fail.
2788 * Return: true if the port uses RoCE GID table mechanism in order to manage
2791 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2794 return rdma_protocol_roce(device, port_num) &&
2795 device->add_gid && device->del_gid;
2799 * Check if the device supports READ W/ INVALIDATE.
2801 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2804 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2805 * has support for it yet.
2807 return rdma_protocol_iwarp(dev, port_num);
2810 int ib_query_gid(struct ib_device *device,
2811 u8 port_num, int index, union ib_gid *gid,
2812 struct ib_gid_attr *attr);
2814 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2816 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2817 struct ifla_vf_info *info);
2818 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2819 struct ifla_vf_stats *stats);
2820 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2823 int ib_query_pkey(struct ib_device *device,
2824 u8 port_num, u16 index, u16 *pkey);
2826 int ib_modify_device(struct ib_device *device,
2827 int device_modify_mask,
2828 struct ib_device_modify *device_modify);
2830 int ib_modify_port(struct ib_device *device,
2831 u8 port_num, int port_modify_mask,
2832 struct ib_port_modify *port_modify);
2834 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2835 struct net_device *ndev, u8 *port_num, u16 *index);
2837 int ib_find_pkey(struct ib_device *device,
2838 u8 port_num, u16 pkey, u16 *index);
2842 * Create a memory registration for all memory in the system and place
2843 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2844 * ULPs to avoid the overhead of dynamic MRs.
2846 * This flag is generally considered unsafe and must only be used in
2847 * extremly trusted environments. Every use of it will log a warning
2848 * in the kernel log.
2850 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2853 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2854 const char *caller);
2855 #define ib_alloc_pd(device, flags) \
2856 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
2857 void ib_dealloc_pd(struct ib_pd *pd);
2860 * rdma_create_ah - Creates an address handle for the given address vector.
2861 * @pd: The protection domain associated with the address handle.
2862 * @ah_attr: The attributes of the address vector.
2864 * The address handle is used to reference a local or global destination
2865 * in all UD QP post sends.
2867 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2870 * rdma_create_user_ah - Creates an address handle for the given address vector.
2871 * It resolves destination mac address for ah attribute of RoCE type.
2872 * @pd: The protection domain associated with the address handle.
2873 * @ah_attr: The attributes of the address vector.
2874 * @udata: pointer to user's input output buffer information need by
2877 * It returns 0 on success and returns appropriate error code on error.
2878 * The address handle is used to reference a local or global destination
2879 * in all UD QP post sends.
2881 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
2882 struct rdma_ah_attr *ah_attr,
2883 struct ib_udata *udata);
2885 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2887 * @hdr: the L3 header to parse
2888 * @net_type: type of header to parse
2889 * @sgid: place to store source gid
2890 * @dgid: place to store destination gid
2892 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2893 enum rdma_network_type net_type,
2894 union ib_gid *sgid, union ib_gid *dgid);
2897 * ib_get_rdma_header_version - Get the header version
2898 * @hdr: the L3 header to parse
2900 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2903 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
2905 * @device: Device on which the received message arrived.
2906 * @port_num: Port on which the received message arrived.
2907 * @wc: Work completion associated with the received message.
2908 * @grh: References the received global route header. This parameter is
2909 * ignored unless the work completion indicates that the GRH is valid.
2910 * @ah_attr: Returned attributes that can be used when creating an address
2911 * handle for replying to the message.
2913 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
2914 const struct ib_wc *wc, const struct ib_grh *grh,
2915 struct rdma_ah_attr *ah_attr);
2918 * ib_create_ah_from_wc - Creates an address handle associated with the
2919 * sender of the specified work completion.
2920 * @pd: The protection domain associated with the address handle.
2921 * @wc: Work completion information associated with a received message.
2922 * @grh: References the received global route header. This parameter is
2923 * ignored unless the work completion indicates that the GRH is valid.
2924 * @port_num: The outbound port number to associate with the address.
2926 * The address handle is used to reference a local or global destination
2927 * in all UD QP post sends.
2929 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2930 const struct ib_grh *grh, u8 port_num);
2933 * rdma_modify_ah - Modifies the address vector associated with an address
2935 * @ah: The address handle to modify.
2936 * @ah_attr: The new address vector attributes to associate with the
2939 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2942 * rdma_query_ah - Queries the address vector associated with an address
2944 * @ah: The address handle to query.
2945 * @ah_attr: The address vector attributes associated with the address
2948 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2951 * rdma_destroy_ah - Destroys an address handle.
2952 * @ah: The address handle to destroy.
2954 int rdma_destroy_ah(struct ib_ah *ah);
2957 * ib_create_srq - Creates a SRQ associated with the specified protection
2959 * @pd: The protection domain associated with the SRQ.
2960 * @srq_init_attr: A list of initial attributes required to create the
2961 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2962 * the actual capabilities of the created SRQ.
2964 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2965 * requested size of the SRQ, and set to the actual values allocated
2966 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2967 * will always be at least as large as the requested values.
2969 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2970 struct ib_srq_init_attr *srq_init_attr);
2973 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2974 * @srq: The SRQ to modify.
2975 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2976 * the current values of selected SRQ attributes are returned.
2977 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2978 * are being modified.
2980 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2981 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2982 * the number of receives queued drops below the limit.
2984 int ib_modify_srq(struct ib_srq *srq,
2985 struct ib_srq_attr *srq_attr,
2986 enum ib_srq_attr_mask srq_attr_mask);
2989 * ib_query_srq - Returns the attribute list and current values for the
2991 * @srq: The SRQ to query.
2992 * @srq_attr: The attributes of the specified SRQ.
2994 int ib_query_srq(struct ib_srq *srq,
2995 struct ib_srq_attr *srq_attr);
2998 * ib_destroy_srq - Destroys the specified SRQ.
2999 * @srq: The SRQ to destroy.
3001 int ib_destroy_srq(struct ib_srq *srq);
3004 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3005 * @srq: The SRQ to post the work request on.
3006 * @recv_wr: A list of work requests to post on the receive queue.
3007 * @bad_recv_wr: On an immediate failure, this parameter will reference
3008 * the work request that failed to be posted on the QP.
3010 static inline int ib_post_srq_recv(struct ib_srq *srq,
3011 struct ib_recv_wr *recv_wr,
3012 struct ib_recv_wr **bad_recv_wr)
3014 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3018 * ib_create_qp - Creates a QP associated with the specified protection
3020 * @pd: The protection domain associated with the QP.
3021 * @qp_init_attr: A list of initial attributes required to create the
3022 * QP. If QP creation succeeds, then the attributes are updated to
3023 * the actual capabilities of the created QP.
3025 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3026 struct ib_qp_init_attr *qp_init_attr);
3029 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3030 * @qp: The QP to modify.
3031 * @attr: On input, specifies the QP attributes to modify. On output,
3032 * the current values of selected QP attributes are returned.
3033 * @attr_mask: A bit-mask used to specify which attributes of the QP
3034 * are being modified.
3035 * @udata: pointer to user's input output buffer information
3036 * are being modified.
3037 * It returns 0 on success and returns appropriate error code on error.
3039 int ib_modify_qp_with_udata(struct ib_qp *qp,
3040 struct ib_qp_attr *attr,
3042 struct ib_udata *udata);
3045 * ib_modify_qp - Modifies the attributes for the specified QP and then
3046 * transitions the QP to the given state.
3047 * @qp: The QP to modify.
3048 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3049 * the current values of selected QP attributes are returned.
3050 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3051 * are being modified.
3053 int ib_modify_qp(struct ib_qp *qp,
3054 struct ib_qp_attr *qp_attr,
3058 * ib_query_qp - Returns the attribute list and current values for the
3060 * @qp: The QP to query.
3061 * @qp_attr: The attributes of the specified QP.
3062 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3063 * @qp_init_attr: Additional attributes of the selected QP.
3065 * The qp_attr_mask may be used to limit the query to gathering only the
3066 * selected attributes.
3068 int ib_query_qp(struct ib_qp *qp,
3069 struct ib_qp_attr *qp_attr,
3071 struct ib_qp_init_attr *qp_init_attr);
3074 * ib_destroy_qp - Destroys the specified QP.
3075 * @qp: The QP to destroy.
3077 int ib_destroy_qp(struct ib_qp *qp);
3080 * ib_open_qp - Obtain a reference to an existing sharable QP.
3081 * @xrcd - XRC domain
3082 * @qp_open_attr: Attributes identifying the QP to open.
3084 * Returns a reference to a sharable QP.
3086 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3087 struct ib_qp_open_attr *qp_open_attr);
3090 * ib_close_qp - Release an external reference to a QP.
3091 * @qp: The QP handle to release
3093 * The opened QP handle is released by the caller. The underlying
3094 * shared QP is not destroyed until all internal references are released.
3096 int ib_close_qp(struct ib_qp *qp);
3099 * ib_post_send - Posts a list of work requests to the send queue of
3101 * @qp: The QP to post the work request on.
3102 * @send_wr: A list of work requests to post on the send queue.
3103 * @bad_send_wr: On an immediate failure, this parameter will reference
3104 * the work request that failed to be posted on the QP.
3106 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3107 * error is returned, the QP state shall not be affected,
3108 * ib_post_send() will return an immediate error after queueing any
3109 * earlier work requests in the list.
3111 static inline int ib_post_send(struct ib_qp *qp,
3112 struct ib_send_wr *send_wr,
3113 struct ib_send_wr **bad_send_wr)
3115 return qp->device->post_send(qp, send_wr, bad_send_wr);
3119 * ib_post_recv - Posts a list of work requests to the receive queue of
3121 * @qp: The QP to post the work request on.
3122 * @recv_wr: A list of work requests to post on the receive queue.
3123 * @bad_recv_wr: On an immediate failure, this parameter will reference
3124 * the work request that failed to be posted on the QP.
3126 static inline int ib_post_recv(struct ib_qp *qp,
3127 struct ib_recv_wr *recv_wr,
3128 struct ib_recv_wr **bad_recv_wr)
3130 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3133 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3134 int nr_cqe, int comp_vector,
3135 enum ib_poll_context poll_ctx, const char *caller);
3136 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3137 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3139 void ib_free_cq(struct ib_cq *cq);
3140 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3143 * ib_create_cq - Creates a CQ on the specified device.
3144 * @device: The device on which to create the CQ.
3145 * @comp_handler: A user-specified callback that is invoked when a
3146 * completion event occurs on the CQ.
3147 * @event_handler: A user-specified callback that is invoked when an
3148 * asynchronous event not associated with a completion occurs on the CQ.
3149 * @cq_context: Context associated with the CQ returned to the user via
3150 * the associated completion and event handlers.
3151 * @cq_attr: The attributes the CQ should be created upon.
3153 * Users can examine the cq structure to determine the actual CQ size.
3155 struct ib_cq *ib_create_cq(struct ib_device *device,
3156 ib_comp_handler comp_handler,
3157 void (*event_handler)(struct ib_event *, void *),
3159 const struct ib_cq_init_attr *cq_attr);
3162 * ib_resize_cq - Modifies the capacity of the CQ.
3163 * @cq: The CQ to resize.
3164 * @cqe: The minimum size of the CQ.
3166 * Users can examine the cq structure to determine the actual CQ size.
3168 int ib_resize_cq(struct ib_cq *cq, int cqe);
3171 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3172 * @cq: The CQ to modify.
3173 * @cq_count: number of CQEs that will trigger an event
3174 * @cq_period: max period of time in usec before triggering an event
3177 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3180 * ib_destroy_cq - Destroys the specified CQ.
3181 * @cq: The CQ to destroy.
3183 int ib_destroy_cq(struct ib_cq *cq);
3186 * ib_poll_cq - poll a CQ for completion(s)
3187 * @cq:the CQ being polled
3188 * @num_entries:maximum number of completions to return
3189 * @wc:array of at least @num_entries &struct ib_wc where completions
3192 * Poll a CQ for (possibly multiple) completions. If the return value
3193 * is < 0, an error occurred. If the return value is >= 0, it is the
3194 * number of completions returned. If the return value is
3195 * non-negative and < num_entries, then the CQ was emptied.
3197 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3200 return cq->device->poll_cq(cq, num_entries, wc);
3204 * ib_peek_cq - Returns the number of unreaped completions currently
3205 * on the specified CQ.
3206 * @cq: The CQ to peek.
3207 * @wc_cnt: A minimum number of unreaped completions to check for.
3209 * If the number of unreaped completions is greater than or equal to wc_cnt,
3210 * this function returns wc_cnt, otherwise, it returns the actual number of
3211 * unreaped completions.
3213 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3216 * ib_req_notify_cq - Request completion notification on a CQ.
3217 * @cq: The CQ to generate an event for.
3219 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3220 * to request an event on the next solicited event or next work
3221 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3222 * may also be |ed in to request a hint about missed events, as
3226 * < 0 means an error occurred while requesting notification
3227 * == 0 means notification was requested successfully, and if
3228 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3229 * were missed and it is safe to wait for another event. In
3230 * this case is it guaranteed that any work completions added
3231 * to the CQ since the last CQ poll will trigger a completion
3232 * notification event.
3233 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3234 * in. It means that the consumer must poll the CQ again to
3235 * make sure it is empty to avoid missing an event because of a
3236 * race between requesting notification and an entry being
3237 * added to the CQ. This return value means it is possible
3238 * (but not guaranteed) that a work completion has been added
3239 * to the CQ since the last poll without triggering a
3240 * completion notification event.
3242 static inline int ib_req_notify_cq(struct ib_cq *cq,
3243 enum ib_cq_notify_flags flags)
3245 return cq->device->req_notify_cq(cq, flags);
3249 * ib_req_ncomp_notif - Request completion notification when there are
3250 * at least the specified number of unreaped completions on the CQ.
3251 * @cq: The CQ to generate an event for.
3252 * @wc_cnt: The number of unreaped completions that should be on the
3253 * CQ before an event is generated.
3255 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3257 return cq->device->req_ncomp_notif ?
3258 cq->device->req_ncomp_notif(cq, wc_cnt) :
3263 * ib_dma_mapping_error - check a DMA addr for error
3264 * @dev: The device for which the dma_addr was created
3265 * @dma_addr: The DMA address to check
3267 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3269 return dma_mapping_error(dev->dma_device, dma_addr);
3273 * ib_dma_map_single - Map a kernel virtual address to DMA address
3274 * @dev: The device for which the dma_addr is to be created
3275 * @cpu_addr: The kernel virtual address
3276 * @size: The size of the region in bytes
3277 * @direction: The direction of the DMA
3279 static inline u64 ib_dma_map_single(struct ib_device *dev,
3280 void *cpu_addr, size_t size,
3281 enum dma_data_direction direction)
3283 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3287 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3288 * @dev: The device for which the DMA address was created
3289 * @addr: The DMA address
3290 * @size: The size of the region in bytes
3291 * @direction: The direction of the DMA
3293 static inline void ib_dma_unmap_single(struct ib_device *dev,
3294 u64 addr, size_t size,
3295 enum dma_data_direction direction)
3297 dma_unmap_single(dev->dma_device, addr, size, direction);
3301 * ib_dma_map_page - Map a physical page to DMA address
3302 * @dev: The device for which the dma_addr is to be created
3303 * @page: The page to be mapped
3304 * @offset: The offset within the page
3305 * @size: The size of the region in bytes
3306 * @direction: The direction of the DMA
3308 static inline u64 ib_dma_map_page(struct ib_device *dev,
3310 unsigned long offset,
3312 enum dma_data_direction direction)
3314 return dma_map_page(dev->dma_device, page, offset, size, direction);
3318 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3319 * @dev: The device for which the DMA address was created
3320 * @addr: The DMA address
3321 * @size: The size of the region in bytes
3322 * @direction: The direction of the DMA
3324 static inline void ib_dma_unmap_page(struct ib_device *dev,
3325 u64 addr, size_t size,
3326 enum dma_data_direction direction)
3328 dma_unmap_page(dev->dma_device, addr, size, direction);
3332 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3333 * @dev: The device for which the DMA addresses are to be created
3334 * @sg: The array of scatter/gather entries
3335 * @nents: The number of scatter/gather entries
3336 * @direction: The direction of the DMA
3338 static inline int ib_dma_map_sg(struct ib_device *dev,
3339 struct scatterlist *sg, int nents,
3340 enum dma_data_direction direction)
3342 return dma_map_sg(dev->dma_device, sg, nents, direction);
3346 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3347 * @dev: The device for which the DMA addresses were created
3348 * @sg: The array of scatter/gather entries
3349 * @nents: The number of scatter/gather entries
3350 * @direction: The direction of the DMA
3352 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3353 struct scatterlist *sg, int nents,
3354 enum dma_data_direction direction)
3356 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3359 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3360 struct scatterlist *sg, int nents,
3361 enum dma_data_direction direction,
3362 unsigned long dma_attrs)
3364 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3368 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3369 struct scatterlist *sg, int nents,
3370 enum dma_data_direction direction,
3371 unsigned long dma_attrs)
3373 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3376 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3377 * @dev: The device for which the DMA addresses were created
3378 * @sg: The scatter/gather entry
3380 * Note: this function is obsolete. To do: change all occurrences of
3381 * ib_sg_dma_address() into sg_dma_address().
3383 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3384 struct scatterlist *sg)
3386 return sg_dma_address(sg);
3390 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3391 * @dev: The device for which the DMA addresses were created
3392 * @sg: The scatter/gather entry
3394 * Note: this function is obsolete. To do: change all occurrences of
3395 * ib_sg_dma_len() into sg_dma_len().
3397 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3398 struct scatterlist *sg)
3400 return sg_dma_len(sg);
3404 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3405 * @dev: The device for which the DMA address was created
3406 * @addr: The DMA address
3407 * @size: The size of the region in bytes
3408 * @dir: The direction of the DMA
3410 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3413 enum dma_data_direction dir)
3415 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3419 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3420 * @dev: The device for which the DMA address was created
3421 * @addr: The DMA address
3422 * @size: The size of the region in bytes
3423 * @dir: The direction of the DMA
3425 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3428 enum dma_data_direction dir)
3430 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3434 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3435 * @dev: The device for which the DMA address is requested
3436 * @size: The size of the region to allocate in bytes
3437 * @dma_handle: A pointer for returning the DMA address of the region
3438 * @flag: memory allocator flags
3440 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3442 dma_addr_t *dma_handle,
3445 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3449 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3450 * @dev: The device for which the DMA addresses were allocated
3451 * @size: The size of the region
3452 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3453 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3455 static inline void ib_dma_free_coherent(struct ib_device *dev,
3456 size_t size, void *cpu_addr,
3457 dma_addr_t dma_handle)
3459 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3463 * ib_dereg_mr - Deregisters a memory region and removes it from the
3464 * HCA translation table.
3465 * @mr: The memory region to deregister.
3467 * This function can fail, if the memory region has memory windows bound to it.
3469 int ib_dereg_mr(struct ib_mr *mr);
3471 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3472 enum ib_mr_type mr_type,
3476 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3478 * @mr - struct ib_mr pointer to be updated.
3479 * @newkey - new key to be used.
3481 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3483 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3484 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3488 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3489 * for calculating a new rkey for type 2 memory windows.
3490 * @rkey - the rkey to increment.
3492 static inline u32 ib_inc_rkey(u32 rkey)
3494 const u32 mask = 0x000000ff;
3495 return ((rkey + 1) & mask) | (rkey & ~mask);
3499 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3500 * @pd: The protection domain associated with the unmapped region.
3501 * @mr_access_flags: Specifies the memory access rights.
3502 * @fmr_attr: Attributes of the unmapped region.
3504 * A fast memory region must be mapped before it can be used as part of
3507 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3508 int mr_access_flags,
3509 struct ib_fmr_attr *fmr_attr);
3512 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3513 * @fmr: The fast memory region to associate with the pages.
3514 * @page_list: An array of physical pages to map to the fast memory region.
3515 * @list_len: The number of pages in page_list.
3516 * @iova: The I/O virtual address to use with the mapped region.
3518 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3519 u64 *page_list, int list_len,
3522 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3526 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3527 * @fmr_list: A linked list of fast memory regions to unmap.
3529 int ib_unmap_fmr(struct list_head *fmr_list);
3532 * ib_dealloc_fmr - Deallocates a fast memory region.
3533 * @fmr: The fast memory region to deallocate.
3535 int ib_dealloc_fmr(struct ib_fmr *fmr);
3538 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3539 * @qp: QP to attach to the multicast group. The QP must be type
3541 * @gid: Multicast group GID.
3542 * @lid: Multicast group LID in host byte order.
3544 * In order to send and receive multicast packets, subnet
3545 * administration must have created the multicast group and configured
3546 * the fabric appropriately. The port associated with the specified
3547 * QP must also be a member of the multicast group.
3549 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3552 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3553 * @qp: QP to detach from the multicast group.
3554 * @gid: Multicast group GID.
3555 * @lid: Multicast group LID in host byte order.
3557 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3560 * ib_alloc_xrcd - Allocates an XRC domain.
3561 * @device: The device on which to allocate the XRC domain.
3562 * @caller: Module name for kernel consumers
3564 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3565 #define ib_alloc_xrcd(device) \
3566 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3569 * ib_dealloc_xrcd - Deallocates an XRC domain.
3570 * @xrcd: The XRC domain to deallocate.
3572 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3574 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3575 struct ib_flow_attr *flow_attr, int domain);
3576 int ib_destroy_flow(struct ib_flow *flow_id);
3578 static inline int ib_check_mr_access(int flags)
3581 * Local write permission is required if remote write or
3582 * remote atomic permission is also requested.
3584 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3585 !(flags & IB_ACCESS_LOCAL_WRITE))
3592 * ib_check_mr_status: lightweight check of MR status.
3593 * This routine may provide status checks on a selected
3594 * ib_mr. first use is for signature status check.
3596 * @mr: A memory region.
3597 * @check_mask: Bitmask of which checks to perform from
3598 * ib_mr_status_check enumeration.
3599 * @mr_status: The container of relevant status checks.
3600 * failed checks will be indicated in the status bitmask
3601 * and the relevant info shall be in the error item.
3603 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3604 struct ib_mr_status *mr_status);
3606 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3607 u16 pkey, const union ib_gid *gid,
3608 const struct sockaddr *addr);
3609 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3610 struct ib_wq_init_attr *init_attr);
3611 int ib_destroy_wq(struct ib_wq *wq);
3612 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3614 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3615 struct ib_rwq_ind_table_init_attr*
3616 wq_ind_table_init_attr);
3617 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3619 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3620 unsigned int *sg_offset, unsigned int page_size);
3623 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3624 unsigned int *sg_offset, unsigned int page_size)
3628 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3634 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3635 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3637 void ib_drain_rq(struct ib_qp *qp);
3638 void ib_drain_sq(struct ib_qp *qp);
3639 void ib_drain_qp(struct ib_qp *qp);
3641 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3643 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3645 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3646 return attr->roce.dmac;
3650 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3652 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3653 attr->ib.dlid = (u16)dlid;
3654 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3655 attr->opa.dlid = dlid;
3658 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3660 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3661 return attr->ib.dlid;
3662 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3663 return attr->opa.dlid;
3667 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3672 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3677 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3680 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3681 attr->ib.src_path_bits = src_path_bits;
3682 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3683 attr->opa.src_path_bits = src_path_bits;
3686 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3688 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3689 return attr->ib.src_path_bits;
3690 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3691 return attr->opa.src_path_bits;
3695 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3698 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3699 attr->opa.make_grd = make_grd;
3702 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3704 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3705 return attr->opa.make_grd;
3709 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3711 attr->port_num = port_num;
3714 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3716 return attr->port_num;
3719 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3722 attr->static_rate = static_rate;
3725 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3727 return attr->static_rate;
3730 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3731 enum ib_ah_flags flag)
3733 attr->ah_flags = flag;
3736 static inline enum ib_ah_flags
3737 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3739 return attr->ah_flags;
3742 static inline const struct ib_global_route
3743 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3748 /*To retrieve and modify the grh */
3749 static inline struct ib_global_route
3750 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3755 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3757 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3759 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3762 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3765 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3767 grh->dgid.global.subnet_prefix = prefix;
3770 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3773 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3775 grh->dgid.global.interface_id = if_id;
3778 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3779 union ib_gid *dgid, u32 flow_label,
3780 u8 sgid_index, u8 hop_limit,
3783 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3785 attr->ah_flags = IB_AH_GRH;
3788 grh->flow_label = flow_label;
3789 grh->sgid_index = sgid_index;
3790 grh->hop_limit = hop_limit;
3791 grh->traffic_class = traffic_class;
3795 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3798 if (rdma_protocol_roce(dev, port_num))
3799 return RDMA_AH_ATTR_TYPE_ROCE;
3800 else if ((rdma_protocol_ib(dev, port_num)) &&
3801 (rdma_cap_opa_ah(dev, port_num)))
3802 return RDMA_AH_ATTR_TYPE_OPA;
3804 return RDMA_AH_ATTR_TYPE_IB;
3808 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3809 * In the current implementation the only way to get
3810 * get the 32bit lid is from other sources for OPA.
3811 * For IB, lids will always be 16bits so cast the
3812 * value accordingly.
3816 static inline u16 ib_lid_cpu16(u32 lid)
3818 WARN_ON_ONCE(lid & 0xFFFF0000);
3823 * ib_lid_be16 - Return lid in 16bit BE encoding.
3827 static inline __be16 ib_lid_be16(u32 lid)
3829 WARN_ON_ONCE(lid & 0xFFFF0000);
3830 return cpu_to_be16((u16)lid);
3834 * ib_get_vector_affinity - Get the affinity mappings of a given completion
3836 * @device: the rdma device
3837 * @comp_vector: index of completion vector
3839 * Returns NULL on failure, otherwise a corresponding cpu map of the
3840 * completion vector (returns all-cpus map if the device driver doesn't
3841 * implement get_vector_affinity).
3843 static inline const struct cpumask *
3844 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3846 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3847 !device->get_vector_affinity)
3850 return device->get_vector_affinity(device, comp_vector);
3855 * rdma_roce_rescan_device - Rescan all of the network devices in the system
3856 * and add their gids, as needed, to the relevant RoCE devices.
3858 * @device: the rdma device
3860 void rdma_roce_rescan_device(struct ib_device *ibdev);
3862 #endif /* IB_VERBS_H */