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>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <uapi/rdma/ib_user_verbs.h>
63 #include <rdma/restrack.h>
64 #include <uapi/rdma/rdma_user_ioctl.h>
65 #include <uapi/rdma/ib_user_ioctl_verbs.h>
67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
71 extern struct workqueue_struct *ib_wq;
72 extern struct workqueue_struct *ib_comp_wq;
73 extern struct workqueue_struct *ib_comp_unbound_wq;
83 extern union ib_gid zgid;
86 /* If link layer is Ethernet, this is RoCE V1 */
89 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
93 #define ROCE_V2_UDP_DPORT 4791
95 struct net_device *ndev;
96 struct ib_device *device;
98 enum ib_gid_type gid_type;
103 enum rdma_node_type {
104 /* IB values map to NodeInfo:NodeType. */
114 /* set the local administered indication */
115 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
118 enum rdma_transport_type {
120 RDMA_TRANSPORT_IWARP,
121 RDMA_TRANSPORT_USNIC,
122 RDMA_TRANSPORT_USNIC_UDP
125 enum rdma_protocol_type {
129 RDMA_PROTOCOL_USNIC_UDP
132 __attribute_const__ enum rdma_transport_type
133 rdma_node_get_transport(enum rdma_node_type node_type);
135 enum rdma_network_type {
137 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
144 if (network_type == RDMA_NETWORK_IPV4 ||
145 network_type == RDMA_NETWORK_IPV6)
146 return IB_GID_TYPE_ROCE_UDP_ENCAP;
148 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
149 return IB_GID_TYPE_IB;
152 static inline enum rdma_network_type
153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
155 if (attr->gid_type == IB_GID_TYPE_IB)
156 return RDMA_NETWORK_IB;
158 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
159 return RDMA_NETWORK_IPV4;
161 return RDMA_NETWORK_IPV6;
164 enum rdma_link_layer {
165 IB_LINK_LAYER_UNSPECIFIED,
166 IB_LINK_LAYER_INFINIBAND,
167 IB_LINK_LAYER_ETHERNET,
170 enum ib_device_cap_flags {
171 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
172 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
173 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
174 IB_DEVICE_RAW_MULTI = (1 << 3),
175 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
176 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
177 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
178 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
179 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
180 /* Not in use, former INIT_TYPE = (1 << 9),*/
181 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
182 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
183 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
184 IB_DEVICE_SRQ_RESIZE = (1 << 13),
185 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
188 * This device supports a per-device lkey or stag that can be
189 * used without performing a memory registration for the local
190 * memory. Note that ULPs should never check this flag, but
191 * instead of use the local_dma_lkey flag in the ib_pd structure,
192 * which will always contain a usable lkey.
194 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
195 /* Reserved, old SEND_W_INV = (1 << 16),*/
196 IB_DEVICE_MEM_WINDOW = (1 << 17),
198 * Devices should set IB_DEVICE_UD_IP_SUM if they support
199 * insertion of UDP and TCP checksum on outgoing UD IPoIB
200 * messages and can verify the validity of checksum for
201 * incoming messages. Setting this flag implies that the
202 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
204 IB_DEVICE_UD_IP_CSUM = (1 << 18),
205 IB_DEVICE_UD_TSO = (1 << 19),
206 IB_DEVICE_XRC = (1 << 20),
209 * This device supports the IB "base memory management extension",
210 * which includes support for fast registrations (IB_WR_REG_MR,
211 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
212 * also be set by any iWarp device which must support FRs to comply
213 * to the iWarp verbs spec. iWarp devices also support the
214 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
217 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
218 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
219 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
220 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
221 IB_DEVICE_RC_IP_CSUM = (1 << 25),
222 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
223 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
225 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
226 * support execution of WQEs that involve synchronization
227 * of I/O operations with single completion queue managed
230 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
231 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
232 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
233 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
234 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
235 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
236 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
237 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
238 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
239 /* The device supports padding incoming writes to cacheline. */
240 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
243 enum ib_signature_prot_cap {
244 IB_PROT_T10DIF_TYPE_1 = 1,
245 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
246 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
249 enum ib_signature_guard_cap {
250 IB_GUARD_T10DIF_CRC = 1,
251 IB_GUARD_T10DIF_CSUM = 1 << 1,
260 enum ib_odp_general_cap_bits {
261 IB_ODP_SUPPORT = 1 << 0,
262 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
265 enum ib_odp_transport_cap_bits {
266 IB_ODP_SUPPORT_SEND = 1 << 0,
267 IB_ODP_SUPPORT_RECV = 1 << 1,
268 IB_ODP_SUPPORT_WRITE = 1 << 2,
269 IB_ODP_SUPPORT_READ = 1 << 3,
270 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
274 uint64_t general_caps;
276 uint32_t rc_odp_caps;
277 uint32_t uc_odp_caps;
278 uint32_t ud_odp_caps;
279 } per_transport_caps;
283 /* Corresponding bit will be set if qp type from
284 * 'enum ib_qp_type' is supported, e.g.
285 * supported_qpts |= 1 << IB_QPT_UD
288 u32 max_rwq_indirection_tables;
289 u32 max_rwq_indirection_table_size;
292 enum ib_tm_cap_flags {
293 /* Support tag matching on RC transport */
294 IB_TM_CAP_RC = 1 << 0,
298 /* Max size of RNDV header */
299 u32 max_rndv_hdr_size;
300 /* Max number of entries in tag matching list */
302 /* From enum ib_tm_cap_flags */
304 /* Max number of outstanding list operations */
306 /* Max number of SGE in tag matching entry */
310 struct ib_cq_init_attr {
316 enum ib_cq_attr_mask {
317 IB_CQ_MODERATE = 1 << 0,
321 u16 max_cq_moderation_count;
322 u16 max_cq_moderation_period;
325 struct ib_dm_mr_attr {
331 struct ib_dm_alloc_attr {
337 struct ib_device_attr {
339 __be64 sys_image_guid;
347 u64 device_cap_flags;
358 int max_qp_init_rd_atom;
359 int max_ee_init_rd_atom;
360 enum ib_atomic_cap atomic_cap;
361 enum ib_atomic_cap masked_atomic_cap;
368 int max_mcast_qp_attach;
369 int max_total_mcast_qp_attach;
376 unsigned int max_fast_reg_page_list_len;
378 u8 local_ca_ack_delay;
381 struct ib_odp_caps odp_caps;
382 uint64_t timestamp_mask;
383 uint64_t hca_core_clock; /* in KHZ */
384 struct ib_rss_caps rss_caps;
386 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
387 struct ib_tm_caps tm_caps;
388 struct ib_cq_caps cq_caps;
400 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
403 case IB_MTU_256: return 256;
404 case IB_MTU_512: return 512;
405 case IB_MTU_1024: return 1024;
406 case IB_MTU_2048: return 2048;
407 case IB_MTU_4096: return 4096;
412 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
416 else if (mtu >= 2048)
418 else if (mtu >= 1024)
432 IB_PORT_ACTIVE_DEFER = 5
443 static inline int ib_width_enum_to_int(enum ib_port_width width)
446 case IB_WIDTH_1X: return 1;
447 case IB_WIDTH_2X: return 2;
448 case IB_WIDTH_4X: return 4;
449 case IB_WIDTH_8X: return 8;
450 case IB_WIDTH_12X: return 12;
466 * struct rdma_hw_stats
467 * @lock - Mutex to protect parallel write access to lifespan and values
468 * of counters, which are 64bits and not guaranteeed to be written
469 * atomicaly on 32bits systems.
470 * @timestamp - Used by the core code to track when the last update was
471 * @lifespan - Used by the core code to determine how old the counters
472 * should be before being updated again. Stored in jiffies, defaults
473 * to 10 milliseconds, drivers can override the default be specifying
474 * their own value during their allocation routine.
475 * @name - Array of pointers to static names used for the counters in
477 * @num_counters - How many hardware counters there are. If name is
478 * shorter than this number, a kernel oops will result. Driver authors
479 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
480 * in their code to prevent this.
481 * @value - Array of u64 counters that are accessed by the sysfs code and
482 * filled in by the drivers get_stats routine
484 struct rdma_hw_stats {
485 struct mutex lock; /* Protect lifespan and values[] */
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
534 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
536 /* Protocol 0xFFF00000 */
537 #define RDMA_CORE_CAP_PROT_IB 0x00100000
538 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
539 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
541 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
542 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
544 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
545 | RDMA_CORE_CAP_PROT_ROCE \
546 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
548 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
549 | RDMA_CORE_CAP_IB_MAD \
550 | RDMA_CORE_CAP_IB_SMI \
551 | RDMA_CORE_CAP_IB_CM \
552 | RDMA_CORE_CAP_IB_SA \
553 | RDMA_CORE_CAP_AF_IB)
554 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
555 | RDMA_CORE_CAP_IB_MAD \
556 | RDMA_CORE_CAP_IB_CM \
557 | RDMA_CORE_CAP_AF_IB \
558 | RDMA_CORE_CAP_ETH_AH)
559 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
560 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
561 | RDMA_CORE_CAP_IB_MAD \
562 | RDMA_CORE_CAP_IB_CM \
563 | RDMA_CORE_CAP_AF_IB \
564 | RDMA_CORE_CAP_ETH_AH)
565 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
566 | RDMA_CORE_CAP_IW_CM)
567 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
568 | RDMA_CORE_CAP_OPA_MAD)
570 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
572 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
574 struct ib_port_attr {
576 enum ib_port_state state;
578 enum ib_mtu active_mtu;
580 unsigned int ip_gids:1;
581 /* This is the value from PortInfo CapabilityMask, defined by IBA */
600 enum ib_device_modify_flags {
601 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
602 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
605 #define IB_DEVICE_NODE_DESC_MAX 64
607 struct ib_device_modify {
609 char node_desc[IB_DEVICE_NODE_DESC_MAX];
612 enum ib_port_modify_flags {
613 IB_PORT_SHUTDOWN = 1,
614 IB_PORT_INIT_TYPE = (1<<2),
615 IB_PORT_RESET_QKEY_CNTR = (1<<3),
616 IB_PORT_OPA_MASK_CHG = (1<<4)
619 struct ib_port_modify {
620 u32 set_port_cap_mask;
621 u32 clr_port_cap_mask;
629 IB_EVENT_QP_ACCESS_ERR,
633 IB_EVENT_PATH_MIG_ERR,
634 IB_EVENT_DEVICE_FATAL,
635 IB_EVENT_PORT_ACTIVE,
638 IB_EVENT_PKEY_CHANGE,
641 IB_EVENT_SRQ_LIMIT_REACHED,
642 IB_EVENT_QP_LAST_WQE_REACHED,
643 IB_EVENT_CLIENT_REREGISTER,
648 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
651 struct ib_device *device;
659 enum ib_event_type event;
662 struct ib_event_handler {
663 struct ib_device *device;
664 void (*handler)(struct ib_event_handler *, struct ib_event *);
665 struct list_head list;
668 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
670 (_ptr)->device = _device; \
671 (_ptr)->handler = _handler; \
672 INIT_LIST_HEAD(&(_ptr)->list); \
675 struct ib_global_route {
676 const struct ib_gid_attr *sgid_attr;
685 __be32 version_tclass_flow;
693 union rdma_network_hdr {
696 /* The IB spec states that if it's IPv4, the header
697 * is located in the last 20 bytes of the header.
700 struct iphdr roce4grh;
704 #define IB_QPN_MASK 0xFFFFFF
707 IB_MULTICAST_QPN = 0xffffff
710 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
711 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
718 IB_RATE_PORT_CURRENT = 0,
719 IB_RATE_2_5_GBPS = 2,
727 IB_RATE_120_GBPS = 10,
728 IB_RATE_14_GBPS = 11,
729 IB_RATE_56_GBPS = 12,
730 IB_RATE_112_GBPS = 13,
731 IB_RATE_168_GBPS = 14,
732 IB_RATE_25_GBPS = 15,
733 IB_RATE_100_GBPS = 16,
734 IB_RATE_200_GBPS = 17,
735 IB_RATE_300_GBPS = 18,
736 IB_RATE_28_GBPS = 19,
737 IB_RATE_50_GBPS = 20,
738 IB_RATE_400_GBPS = 21,
739 IB_RATE_600_GBPS = 22,
743 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
744 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
745 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
746 * @rate: rate to convert.
748 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
751 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
752 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
753 * @rate: rate to convert.
755 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
759 * enum ib_mr_type - memory region type
760 * @IB_MR_TYPE_MEM_REG: memory region that is used for
761 * normal registration
762 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
763 * signature operations (data-integrity
765 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
766 * register any arbitrary sg lists (without
767 * the normal mr constraints - see
772 IB_MR_TYPE_SIGNATURE,
778 * IB_SIG_TYPE_NONE: Unprotected.
779 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
781 enum ib_signature_type {
787 * Signature T10-DIF block-guard types
788 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
789 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
791 enum ib_t10_dif_bg_type {
797 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
799 * @bg_type: T10-DIF block guard type (CRC|CSUM)
800 * @pi_interval: protection information interval.
801 * @bg: seed of guard computation.
802 * @app_tag: application tag of guard block
803 * @ref_tag: initial guard block reference tag.
804 * @ref_remap: Indicate wethear the reftag increments each block
805 * @app_escape: Indicate to skip block check if apptag=0xffff
806 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
807 * @apptag_check_mask: check bitmask of application tag.
809 struct ib_t10_dif_domain {
810 enum ib_t10_dif_bg_type bg_type;
818 u16 apptag_check_mask;
822 * struct ib_sig_domain - Parameters for signature domain
823 * @sig_type: specific signauture type
824 * @sig: union of all signature domain attributes that may
825 * be used to set domain layout.
827 struct ib_sig_domain {
828 enum ib_signature_type sig_type;
830 struct ib_t10_dif_domain dif;
835 * struct ib_sig_attrs - Parameters for signature handover operation
836 * @check_mask: bitmask for signature byte check (8 bytes)
837 * @mem: memory domain layout desciptor.
838 * @wire: wire domain layout desciptor.
840 struct ib_sig_attrs {
842 struct ib_sig_domain mem;
843 struct ib_sig_domain wire;
846 enum ib_sig_err_type {
853 * Signature check masks (8 bytes in total) according to the T10-PI standard:
854 * -------- -------- ------------
855 * | GUARD | APPTAG | REFTAG |
857 * -------- -------- ------------
860 IB_SIG_CHECK_GUARD = 0xc0,
861 IB_SIG_CHECK_APPTAG = 0x30,
862 IB_SIG_CHECK_REFTAG = 0x0f,
866 * struct ib_sig_err - signature error descriptor
869 enum ib_sig_err_type err_type;
876 enum ib_mr_status_check {
877 IB_MR_CHECK_SIG_STATUS = 1,
881 * struct ib_mr_status - Memory region status container
883 * @fail_status: Bitmask of MR checks status. For each
884 * failed check a corresponding status bit is set.
885 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
888 struct ib_mr_status {
890 struct ib_sig_err sig_err;
894 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
896 * @mult: multiple to convert.
898 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
900 enum rdma_ah_attr_type {
901 RDMA_AH_ATTR_TYPE_UNDEFINED,
902 RDMA_AH_ATTR_TYPE_IB,
903 RDMA_AH_ATTR_TYPE_ROCE,
904 RDMA_AH_ATTR_TYPE_OPA,
912 struct roce_ah_attr {
922 struct rdma_ah_attr {
923 struct ib_global_route grh;
928 enum rdma_ah_attr_type type;
930 struct ib_ah_attr ib;
931 struct roce_ah_attr roce;
932 struct opa_ah_attr opa;
940 IB_WC_LOC_EEC_OP_ERR,
945 IB_WC_LOC_ACCESS_ERR,
946 IB_WC_REM_INV_REQ_ERR,
947 IB_WC_REM_ACCESS_ERR,
950 IB_WC_RNR_RETRY_EXC_ERR,
951 IB_WC_LOC_RDD_VIOL_ERR,
952 IB_WC_REM_INV_RD_REQ_ERR,
955 IB_WC_INV_EEC_STATE_ERR,
957 IB_WC_RESP_TIMEOUT_ERR,
961 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
972 IB_WC_MASKED_COMP_SWAP,
973 IB_WC_MASKED_FETCH_ADD,
975 * Set value of IB_WC_RECV so consumers can test if a completion is a
976 * receive by testing (opcode & IB_WC_RECV).
979 IB_WC_RECV_RDMA_WITH_IMM
984 IB_WC_WITH_IMM = (1<<1),
985 IB_WC_WITH_INVALIDATE = (1<<2),
986 IB_WC_IP_CSUM_OK = (1<<3),
987 IB_WC_WITH_SMAC = (1<<4),
988 IB_WC_WITH_VLAN = (1<<5),
989 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
995 struct ib_cqe *wr_cqe;
997 enum ib_wc_status status;
998 enum ib_wc_opcode opcode;
1004 u32 invalidate_rkey;
1012 u8 port_num; /* valid only for DR SMPs on switches */
1015 u8 network_hdr_type;
1018 enum ib_cq_notify_flags {
1019 IB_CQ_SOLICITED = 1 << 0,
1020 IB_CQ_NEXT_COMP = 1 << 1,
1021 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1022 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1031 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1033 return srq_type == IB_SRQT_XRC ||
1034 srq_type == IB_SRQT_TM;
1037 enum ib_srq_attr_mask {
1038 IB_SRQ_MAX_WR = 1 << 0,
1039 IB_SRQ_LIMIT = 1 << 1,
1042 struct ib_srq_attr {
1048 struct ib_srq_init_attr {
1049 void (*event_handler)(struct ib_event *, void *);
1051 struct ib_srq_attr attr;
1052 enum ib_srq_type srq_type;
1058 struct ib_xrcd *xrcd;
1073 u32 max_inline_data;
1076 * Maximum number of rdma_rw_ctx structures in flight at a time.
1077 * ib_create_qp() will calculate the right amount of neededed WRs
1078 * and MRs based on this.
1090 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1091 * here (and in that order) since the MAD layer uses them as
1092 * indices into a 2-entry table.
1101 IB_QPT_RAW_ETHERTYPE,
1102 IB_QPT_RAW_PACKET = 8,
1106 IB_QPT_DRIVER = 0xFF,
1107 /* Reserve a range for qp types internal to the low level driver.
1108 * These qp types will not be visible at the IB core layer, so the
1109 * IB_QPT_MAX usages should not be affected in the core layer
1111 IB_QPT_RESERVED1 = 0x1000,
1123 enum ib_qp_create_flags {
1124 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1125 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1126 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1127 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1128 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1129 IB_QP_CREATE_NETIF_QP = 1 << 5,
1130 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1131 /* FREE = 1 << 7, */
1132 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1133 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1134 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1135 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1136 /* reserve bits 26-31 for low level drivers' internal use */
1137 IB_QP_CREATE_RESERVED_START = 1 << 26,
1138 IB_QP_CREATE_RESERVED_END = 1 << 31,
1142 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1143 * callback to destroy the passed in QP.
1146 struct ib_qp_init_attr {
1147 /* Consumer's event_handler callback must not block */
1148 void (*event_handler)(struct ib_event *, void *);
1151 struct ib_cq *send_cq;
1152 struct ib_cq *recv_cq;
1154 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1155 struct ib_qp_cap cap;
1156 enum ib_sig_type sq_sig_type;
1157 enum ib_qp_type qp_type;
1161 * Only needed for special QP types, or when using the RW API.
1164 struct ib_rwq_ind_table *rwq_ind_tbl;
1168 struct ib_qp_open_attr {
1169 void (*event_handler)(struct ib_event *, void *);
1172 enum ib_qp_type qp_type;
1175 enum ib_rnr_timeout {
1176 IB_RNR_TIMER_655_36 = 0,
1177 IB_RNR_TIMER_000_01 = 1,
1178 IB_RNR_TIMER_000_02 = 2,
1179 IB_RNR_TIMER_000_03 = 3,
1180 IB_RNR_TIMER_000_04 = 4,
1181 IB_RNR_TIMER_000_06 = 5,
1182 IB_RNR_TIMER_000_08 = 6,
1183 IB_RNR_TIMER_000_12 = 7,
1184 IB_RNR_TIMER_000_16 = 8,
1185 IB_RNR_TIMER_000_24 = 9,
1186 IB_RNR_TIMER_000_32 = 10,
1187 IB_RNR_TIMER_000_48 = 11,
1188 IB_RNR_TIMER_000_64 = 12,
1189 IB_RNR_TIMER_000_96 = 13,
1190 IB_RNR_TIMER_001_28 = 14,
1191 IB_RNR_TIMER_001_92 = 15,
1192 IB_RNR_TIMER_002_56 = 16,
1193 IB_RNR_TIMER_003_84 = 17,
1194 IB_RNR_TIMER_005_12 = 18,
1195 IB_RNR_TIMER_007_68 = 19,
1196 IB_RNR_TIMER_010_24 = 20,
1197 IB_RNR_TIMER_015_36 = 21,
1198 IB_RNR_TIMER_020_48 = 22,
1199 IB_RNR_TIMER_030_72 = 23,
1200 IB_RNR_TIMER_040_96 = 24,
1201 IB_RNR_TIMER_061_44 = 25,
1202 IB_RNR_TIMER_081_92 = 26,
1203 IB_RNR_TIMER_122_88 = 27,
1204 IB_RNR_TIMER_163_84 = 28,
1205 IB_RNR_TIMER_245_76 = 29,
1206 IB_RNR_TIMER_327_68 = 30,
1207 IB_RNR_TIMER_491_52 = 31
1210 enum ib_qp_attr_mask {
1212 IB_QP_CUR_STATE = (1<<1),
1213 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1214 IB_QP_ACCESS_FLAGS = (1<<3),
1215 IB_QP_PKEY_INDEX = (1<<4),
1216 IB_QP_PORT = (1<<5),
1217 IB_QP_QKEY = (1<<6),
1219 IB_QP_PATH_MTU = (1<<8),
1220 IB_QP_TIMEOUT = (1<<9),
1221 IB_QP_RETRY_CNT = (1<<10),
1222 IB_QP_RNR_RETRY = (1<<11),
1223 IB_QP_RQ_PSN = (1<<12),
1224 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1225 IB_QP_ALT_PATH = (1<<14),
1226 IB_QP_MIN_RNR_TIMER = (1<<15),
1227 IB_QP_SQ_PSN = (1<<16),
1228 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1229 IB_QP_PATH_MIG_STATE = (1<<18),
1230 IB_QP_CAP = (1<<19),
1231 IB_QP_DEST_QPN = (1<<20),
1232 IB_QP_RESERVED1 = (1<<21),
1233 IB_QP_RESERVED2 = (1<<22),
1234 IB_QP_RESERVED3 = (1<<23),
1235 IB_QP_RESERVED4 = (1<<24),
1236 IB_QP_RATE_LIMIT = (1<<25),
1261 enum ib_qp_state qp_state;
1262 enum ib_qp_state cur_qp_state;
1263 enum ib_mtu path_mtu;
1264 enum ib_mig_state path_mig_state;
1269 int qp_access_flags;
1270 struct ib_qp_cap cap;
1271 struct rdma_ah_attr ah_attr;
1272 struct rdma_ah_attr alt_ah_attr;
1275 u8 en_sqd_async_notify;
1278 u8 max_dest_rd_atomic;
1290 /* These are shared with userspace */
1291 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1292 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1293 IB_WR_SEND = IB_UVERBS_WR_SEND,
1294 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1295 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1296 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1297 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1298 IB_WR_LSO = IB_UVERBS_WR_TSO,
1299 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1300 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1301 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1302 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1303 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1304 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1305 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1307 /* These are kernel only and can not be issued by userspace */
1308 IB_WR_REG_MR = 0x20,
1311 /* reserve values for low level drivers' internal use.
1312 * These values will not be used at all in the ib core layer.
1314 IB_WR_RESERVED1 = 0xf0,
1326 enum ib_send_flags {
1328 IB_SEND_SIGNALED = (1<<1),
1329 IB_SEND_SOLICITED = (1<<2),
1330 IB_SEND_INLINE = (1<<3),
1331 IB_SEND_IP_CSUM = (1<<4),
1333 /* reserve bits 26-31 for low level drivers' internal use */
1334 IB_SEND_RESERVED_START = (1 << 26),
1335 IB_SEND_RESERVED_END = (1 << 31),
1345 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1349 struct ib_send_wr *next;
1352 struct ib_cqe *wr_cqe;
1354 struct ib_sge *sg_list;
1356 enum ib_wr_opcode opcode;
1360 u32 invalidate_rkey;
1365 struct ib_send_wr wr;
1370 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1372 return container_of(wr, struct ib_rdma_wr, wr);
1375 struct ib_atomic_wr {
1376 struct ib_send_wr wr;
1380 u64 compare_add_mask;
1385 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1387 return container_of(wr, struct ib_atomic_wr, wr);
1391 struct ib_send_wr wr;
1398 u16 pkey_index; /* valid for GSI only */
1399 u8 port_num; /* valid for DR SMPs on switch only */
1402 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1404 return container_of(wr, struct ib_ud_wr, wr);
1408 struct ib_send_wr wr;
1414 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1416 return container_of(wr, struct ib_reg_wr, wr);
1419 struct ib_sig_handover_wr {
1420 struct ib_send_wr wr;
1421 struct ib_sig_attrs *sig_attrs;
1422 struct ib_mr *sig_mr;
1424 struct ib_sge *prot;
1427 static inline const struct ib_sig_handover_wr *
1428 sig_handover_wr(const struct ib_send_wr *wr)
1430 return container_of(wr, struct ib_sig_handover_wr, wr);
1434 struct ib_recv_wr *next;
1437 struct ib_cqe *wr_cqe;
1439 struct ib_sge *sg_list;
1443 enum ib_access_flags {
1444 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1445 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1446 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1447 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1448 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1449 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1450 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1451 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1453 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1457 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1458 * are hidden here instead of a uapi header!
1460 enum ib_mr_rereg_flags {
1461 IB_MR_REREG_TRANS = 1,
1462 IB_MR_REREG_PD = (1<<1),
1463 IB_MR_REREG_ACCESS = (1<<2),
1464 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1467 struct ib_fmr_attr {
1475 enum rdma_remove_reason {
1477 * Userspace requested uobject deletion or initial try
1478 * to remove uobject via cleanup. Call could fail
1480 RDMA_REMOVE_DESTROY,
1481 /* Context deletion. This call should delete the actual object itself */
1483 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1484 RDMA_REMOVE_DRIVER_REMOVE,
1485 /* uobj is being cleaned-up before being committed */
1489 struct ib_rdmacg_object {
1490 #ifdef CONFIG_CGROUP_RDMA
1491 struct rdma_cgroup *cg; /* owner rdma cgroup */
1495 struct ib_ucontext {
1496 struct ib_device *device;
1497 struct ib_uverbs_file *ufile;
1499 * 'closing' can be read by the driver only during a destroy callback,
1500 * it is set when we are closing the file descriptor and indicates
1501 * that mm_sem may be locked.
1505 bool cleanup_retryable;
1507 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1508 void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1509 unsigned long start, unsigned long end);
1510 struct mutex per_mm_list_lock;
1511 struct list_head per_mm_list;
1514 struct ib_rdmacg_object cg_obj;
1516 * Implementation details of the RDMA core, don't use in drivers:
1518 struct rdma_restrack_entry res;
1522 u64 user_handle; /* handle given to us by userspace */
1523 /* ufile & ucontext owning this object */
1524 struct ib_uverbs_file *ufile;
1525 /* FIXME, save memory: ufile->context == context */
1526 struct ib_ucontext *context; /* associated user context */
1527 void *object; /* containing object */
1528 struct list_head list; /* link to context's list */
1529 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1530 int id; /* index into kernel idr */
1532 atomic_t usecnt; /* protects exclusive access */
1533 struct rcu_head rcu; /* kfree_rcu() overhead */
1535 const struct uverbs_api_object *uapi_object;
1539 const void __user *inbuf;
1540 void __user *outbuf;
1548 struct ib_device *device;
1549 struct ib_uobject *uobject;
1550 atomic_t usecnt; /* count all resources */
1552 u32 unsafe_global_rkey;
1555 * Implementation details of the RDMA core, don't use in drivers:
1557 struct ib_mr *__internal_mr;
1558 struct rdma_restrack_entry res;
1562 struct ib_device *device;
1563 atomic_t usecnt; /* count all exposed resources */
1564 struct inode *inode;
1566 struct mutex tgt_qp_mutex;
1567 struct list_head tgt_qp_list;
1571 struct ib_device *device;
1573 struct ib_uobject *uobject;
1574 const struct ib_gid_attr *sgid_attr;
1575 enum rdma_ah_attr_type type;
1578 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1580 enum ib_poll_context {
1581 IB_POLL_DIRECT, /* caller context, no hw completions */
1582 IB_POLL_SOFTIRQ, /* poll from softirq context */
1583 IB_POLL_WORKQUEUE, /* poll from workqueue */
1584 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1588 struct ib_device *device;
1589 struct ib_uobject *uobject;
1590 ib_comp_handler comp_handler;
1591 void (*event_handler)(struct ib_event *, void *);
1594 atomic_t usecnt; /* count number of work queues */
1595 enum ib_poll_context poll_ctx;
1598 struct irq_poll iop;
1599 struct work_struct work;
1601 struct workqueue_struct *comp_wq;
1603 * Implementation details of the RDMA core, don't use in drivers:
1605 struct rdma_restrack_entry res;
1609 struct ib_device *device;
1611 struct ib_uobject *uobject;
1612 void (*event_handler)(struct ib_event *, void *);
1614 enum ib_srq_type srq_type;
1621 struct ib_xrcd *xrcd;
1628 enum ib_raw_packet_caps {
1629 /* Strip cvlan from incoming packet and report it in the matching work
1630 * completion is supported.
1632 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1633 /* Scatter FCS field of an incoming packet to host memory is supported.
1635 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1636 /* Checksum offloads are supported (for both send and receive). */
1637 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1638 /* When a packet is received for an RQ with no receive WQEs, the
1639 * packet processing is delayed.
1641 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1655 struct ib_device *device;
1656 struct ib_uobject *uobject;
1658 void (*event_handler)(struct ib_event *, void *);
1662 enum ib_wq_state state;
1663 enum ib_wq_type wq_type;
1668 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1669 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1670 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1671 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1674 struct ib_wq_init_attr {
1676 enum ib_wq_type wq_type;
1680 void (*event_handler)(struct ib_event *, void *);
1681 u32 create_flags; /* Use enum ib_wq_flags */
1684 enum ib_wq_attr_mask {
1685 IB_WQ_STATE = 1 << 0,
1686 IB_WQ_CUR_STATE = 1 << 1,
1687 IB_WQ_FLAGS = 1 << 2,
1691 enum ib_wq_state wq_state;
1692 enum ib_wq_state curr_wq_state;
1693 u32 flags; /* Use enum ib_wq_flags */
1694 u32 flags_mask; /* Use enum ib_wq_flags */
1697 struct ib_rwq_ind_table {
1698 struct ib_device *device;
1699 struct ib_uobject *uobject;
1702 u32 log_ind_tbl_size;
1703 struct ib_wq **ind_tbl;
1706 struct ib_rwq_ind_table_init_attr {
1707 u32 log_ind_tbl_size;
1708 /* Each entry is a pointer to Receive Work Queue */
1709 struct ib_wq **ind_tbl;
1712 enum port_pkey_state {
1713 IB_PORT_PKEY_NOT_VALID = 0,
1714 IB_PORT_PKEY_VALID = 1,
1715 IB_PORT_PKEY_LISTED = 2,
1718 struct ib_qp_security;
1720 struct ib_port_pkey {
1721 enum port_pkey_state state;
1724 struct list_head qp_list;
1725 struct list_head to_error_list;
1726 struct ib_qp_security *sec;
1729 struct ib_ports_pkeys {
1730 struct ib_port_pkey main;
1731 struct ib_port_pkey alt;
1734 struct ib_qp_security {
1736 struct ib_device *dev;
1737 /* Hold this mutex when changing port and pkey settings. */
1739 struct ib_ports_pkeys *ports_pkeys;
1740 /* A list of all open shared QP handles. Required to enforce security
1741 * properly for all users of a shared QP.
1743 struct list_head shared_qp_list;
1746 atomic_t error_list_count;
1747 struct completion error_complete;
1748 int error_comps_pending;
1752 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1753 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1756 struct ib_device *device;
1758 struct ib_cq *send_cq;
1759 struct ib_cq *recv_cq;
1762 struct list_head rdma_mrs;
1763 struct list_head sig_mrs;
1765 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1766 struct list_head xrcd_list;
1768 /* count times opened, mcast attaches, flow attaches */
1770 struct list_head open_list;
1771 struct ib_qp *real_qp;
1772 struct ib_uobject *uobject;
1773 void (*event_handler)(struct ib_event *, void *);
1775 /* sgid_attrs associated with the AV's */
1776 const struct ib_gid_attr *av_sgid_attr;
1777 const struct ib_gid_attr *alt_path_sgid_attr;
1781 enum ib_qp_type qp_type;
1782 struct ib_rwq_ind_table *rwq_ind_tbl;
1783 struct ib_qp_security *qp_sec;
1787 * Implementation details of the RDMA core, don't use in drivers:
1789 struct rdma_restrack_entry res;
1793 struct ib_device *device;
1796 struct ib_uobject *uobject;
1801 struct ib_device *device;
1807 unsigned int page_size;
1810 struct ib_uobject *uobject; /* user */
1811 struct list_head qp_entry; /* FR */
1817 * Implementation details of the RDMA core, don't use in drivers:
1819 struct rdma_restrack_entry res;
1823 struct ib_device *device;
1825 struct ib_uobject *uobject;
1827 enum ib_mw_type type;
1831 struct ib_device *device;
1833 struct list_head list;
1838 /* Supported steering options */
1839 enum ib_flow_attr_type {
1840 /* steering according to rule specifications */
1841 IB_FLOW_ATTR_NORMAL = 0x0,
1842 /* default unicast and multicast rule -
1843 * receive all Eth traffic which isn't steered to any QP
1845 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1846 /* default multicast rule -
1847 * receive all Eth multicast traffic which isn't steered to any QP
1849 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1850 /* sniffer rule - receive all port traffic */
1851 IB_FLOW_ATTR_SNIFFER = 0x3
1854 /* Supported steering header types */
1855 enum ib_flow_spec_type {
1857 IB_FLOW_SPEC_ETH = 0x20,
1858 IB_FLOW_SPEC_IB = 0x22,
1860 IB_FLOW_SPEC_IPV4 = 0x30,
1861 IB_FLOW_SPEC_IPV6 = 0x31,
1862 IB_FLOW_SPEC_ESP = 0x34,
1864 IB_FLOW_SPEC_TCP = 0x40,
1865 IB_FLOW_SPEC_UDP = 0x41,
1866 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1867 IB_FLOW_SPEC_GRE = 0x51,
1868 IB_FLOW_SPEC_MPLS = 0x60,
1869 IB_FLOW_SPEC_INNER = 0x100,
1871 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1872 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1873 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1874 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1876 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1877 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1879 /* Flow steering rule priority is set according to it's domain.
1880 * Lower domain value means higher priority.
1882 enum ib_flow_domain {
1883 IB_FLOW_DOMAIN_USER,
1884 IB_FLOW_DOMAIN_ETHTOOL,
1887 IB_FLOW_DOMAIN_NUM /* Must be last */
1890 enum ib_flow_flags {
1891 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1892 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1893 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1896 struct ib_flow_eth_filter {
1905 struct ib_flow_spec_eth {
1908 struct ib_flow_eth_filter val;
1909 struct ib_flow_eth_filter mask;
1912 struct ib_flow_ib_filter {
1919 struct ib_flow_spec_ib {
1922 struct ib_flow_ib_filter val;
1923 struct ib_flow_ib_filter mask;
1926 /* IPv4 header flags */
1927 enum ib_ipv4_flags {
1928 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1929 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1930 last have this flag set */
1933 struct ib_flow_ipv4_filter {
1944 struct ib_flow_spec_ipv4 {
1947 struct ib_flow_ipv4_filter val;
1948 struct ib_flow_ipv4_filter mask;
1951 struct ib_flow_ipv6_filter {
1962 struct ib_flow_spec_ipv6 {
1965 struct ib_flow_ipv6_filter val;
1966 struct ib_flow_ipv6_filter mask;
1969 struct ib_flow_tcp_udp_filter {
1976 struct ib_flow_spec_tcp_udp {
1979 struct ib_flow_tcp_udp_filter val;
1980 struct ib_flow_tcp_udp_filter mask;
1983 struct ib_flow_tunnel_filter {
1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1989 * the tunnel_id from val has the vni value
1991 struct ib_flow_spec_tunnel {
1994 struct ib_flow_tunnel_filter val;
1995 struct ib_flow_tunnel_filter mask;
1998 struct ib_flow_esp_filter {
2005 struct ib_flow_spec_esp {
2008 struct ib_flow_esp_filter val;
2009 struct ib_flow_esp_filter mask;
2012 struct ib_flow_gre_filter {
2013 __be16 c_ks_res0_ver;
2020 struct ib_flow_spec_gre {
2023 struct ib_flow_gre_filter val;
2024 struct ib_flow_gre_filter mask;
2027 struct ib_flow_mpls_filter {
2033 struct ib_flow_spec_mpls {
2036 struct ib_flow_mpls_filter val;
2037 struct ib_flow_mpls_filter mask;
2040 struct ib_flow_spec_action_tag {
2041 enum ib_flow_spec_type type;
2046 struct ib_flow_spec_action_drop {
2047 enum ib_flow_spec_type type;
2051 struct ib_flow_spec_action_handle {
2052 enum ib_flow_spec_type type;
2054 struct ib_flow_action *act;
2057 enum ib_counters_description {
2062 struct ib_flow_spec_action_count {
2063 enum ib_flow_spec_type type;
2065 struct ib_counters *counters;
2068 union ib_flow_spec {
2073 struct ib_flow_spec_eth eth;
2074 struct ib_flow_spec_ib ib;
2075 struct ib_flow_spec_ipv4 ipv4;
2076 struct ib_flow_spec_tcp_udp tcp_udp;
2077 struct ib_flow_spec_ipv6 ipv6;
2078 struct ib_flow_spec_tunnel tunnel;
2079 struct ib_flow_spec_esp esp;
2080 struct ib_flow_spec_gre gre;
2081 struct ib_flow_spec_mpls mpls;
2082 struct ib_flow_spec_action_tag flow_tag;
2083 struct ib_flow_spec_action_drop drop;
2084 struct ib_flow_spec_action_handle action;
2085 struct ib_flow_spec_action_count flow_count;
2088 struct ib_flow_attr {
2089 enum ib_flow_attr_type type;
2095 union ib_flow_spec flows[];
2100 struct ib_device *device;
2101 struct ib_uobject *uobject;
2104 enum ib_flow_action_type {
2105 IB_FLOW_ACTION_UNSPECIFIED,
2106 IB_FLOW_ACTION_ESP = 1,
2109 struct ib_flow_action_attrs_esp_keymats {
2110 enum ib_uverbs_flow_action_esp_keymat protocol;
2112 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2116 struct ib_flow_action_attrs_esp_replays {
2117 enum ib_uverbs_flow_action_esp_replay protocol;
2119 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2123 enum ib_flow_action_attrs_esp_flags {
2124 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2125 * This is done in order to share the same flags between user-space and
2126 * kernel and spare an unnecessary translation.
2130 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2131 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2134 struct ib_flow_spec_list {
2135 struct ib_flow_spec_list *next;
2136 union ib_flow_spec spec;
2139 struct ib_flow_action_attrs_esp {
2140 struct ib_flow_action_attrs_esp_keymats *keymat;
2141 struct ib_flow_action_attrs_esp_replays *replay;
2142 struct ib_flow_spec_list *encap;
2143 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2144 * Value of 0 is a valid value.
2150 /* Use enum ib_flow_action_attrs_esp_flags */
2152 u64 hard_limit_pkts;
2155 struct ib_flow_action {
2156 struct ib_device *device;
2157 struct ib_uobject *uobject;
2158 enum ib_flow_action_type type;
2165 enum ib_process_mad_flags {
2166 IB_MAD_IGNORE_MKEY = 1,
2167 IB_MAD_IGNORE_BKEY = 2,
2168 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2171 enum ib_mad_result {
2172 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2173 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2174 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2175 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2178 struct ib_port_cache {
2180 struct ib_pkey_cache *pkey;
2181 struct ib_gid_table *gid;
2183 enum ib_port_state port_state;
2188 struct ib_event_handler event_handler;
2189 struct ib_port_cache *ports;
2194 struct ib_port_immutable {
2201 /* rdma netdev type - specifies protocol type */
2202 enum rdma_netdev_t {
2203 RDMA_NETDEV_OPA_VNIC,
2208 * struct rdma_netdev - rdma netdev
2209 * For cases where netstack interfacing is required.
2211 struct rdma_netdev {
2213 struct ib_device *hca;
2217 * cleanup function must be specified.
2218 * FIXME: This is only used for OPA_VNIC and that usage should be
2221 void (*free_rdma_netdev)(struct net_device *netdev);
2223 /* control functions */
2224 void (*set_id)(struct net_device *netdev, int id);
2226 int (*send)(struct net_device *dev, struct sk_buff *skb,
2227 struct ib_ah *address, u32 dqpn);
2229 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2230 union ib_gid *gid, u16 mlid,
2231 int set_qkey, u32 qkey);
2232 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2233 union ib_gid *gid, u16 mlid);
2236 struct rdma_netdev_alloc_params {
2242 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2243 struct net_device *netdev, void *param);
2246 struct ib_port_pkey_list {
2247 /* Lock to hold while modifying the list. */
2248 spinlock_t list_lock;
2249 struct list_head pkey_list;
2252 struct ib_counters {
2253 struct ib_device *device;
2254 struct ib_uobject *uobject;
2255 /* num of objects attached */
2259 struct ib_counters_read_attr {
2262 u32 flags; /* use enum ib_read_counters_flags */
2265 struct uverbs_attr_bundle;
2268 * struct ib_device_ops - InfiniBand device operations
2269 * This structure defines all the InfiniBand device operations, providers will
2270 * need to define the supported operations, otherwise they will be set to null.
2272 struct ib_device_ops {
2273 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2274 const struct ib_send_wr **bad_send_wr);
2275 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2276 const struct ib_recv_wr **bad_recv_wr);
2277 void (*drain_rq)(struct ib_qp *qp);
2278 void (*drain_sq)(struct ib_qp *qp);
2279 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2280 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2281 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2282 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2283 int (*post_srq_recv)(struct ib_srq *srq,
2284 const struct ib_recv_wr *recv_wr,
2285 const struct ib_recv_wr **bad_recv_wr);
2286 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2287 u8 port_num, const struct ib_wc *in_wc,
2288 const struct ib_grh *in_grh,
2289 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2290 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2291 u16 *out_mad_pkey_index);
2292 int (*query_device)(struct ib_device *device,
2293 struct ib_device_attr *device_attr,
2294 struct ib_udata *udata);
2295 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2296 struct ib_device_modify *device_modify);
2297 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2298 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2300 int (*query_port)(struct ib_device *device, u8 port_num,
2301 struct ib_port_attr *port_attr);
2302 int (*modify_port)(struct ib_device *device, u8 port_num,
2303 int port_modify_mask,
2304 struct ib_port_modify *port_modify);
2306 * The following mandatory functions are used only at device
2307 * registration. Keep functions such as these at the end of this
2308 * structure to avoid cache line misses when accessing struct ib_device
2311 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2312 struct ib_port_immutable *immutable);
2313 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2316 * When calling get_netdev, the HW vendor's driver should return the
2317 * net device of device @device at port @port_num or NULL if such
2318 * a net device doesn't exist. The vendor driver should call dev_hold
2319 * on this net device. The HW vendor's device driver must guarantee
2320 * that this function returns NULL before the net device has finished
2321 * NETDEV_UNREGISTER state.
2323 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2325 * rdma netdev operation
2327 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2328 * must return -EOPNOTSUPP if it doesn't support the specified type.
2330 struct net_device *(*alloc_rdma_netdev)(
2331 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2332 const char *name, unsigned char name_assign_type,
2333 void (*setup)(struct net_device *));
2335 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2336 enum rdma_netdev_t type,
2337 struct rdma_netdev_alloc_params *params);
2339 * query_gid should be return GID value for @device, when @port_num
2340 * link layer is either IB or iWarp. It is no-op if @port_num port
2341 * is RoCE link layer.
2343 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2346 * When calling add_gid, the HW vendor's driver should add the gid
2347 * of device of port at gid index available at @attr. Meta-info of
2348 * that gid (for example, the network device related to this gid) is
2349 * available at @attr. @context allows the HW vendor driver to store
2350 * extra information together with a GID entry. The HW vendor driver may
2351 * allocate memory to contain this information and store it in @context
2352 * when a new GID entry is written to. Params are consistent until the
2353 * next call of add_gid or delete_gid. The function should return 0 on
2354 * success or error otherwise. The function could be called
2355 * concurrently for different ports. This function is only called when
2356 * roce_gid_table is used.
2358 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2360 * When calling del_gid, the HW vendor's driver should delete the
2361 * gid of device @device at gid index gid_index of port port_num
2362 * available in @attr.
2363 * Upon the deletion of a GID entry, the HW vendor must free any
2364 * allocated memory. The caller will clear @context afterwards.
2365 * This function is only called when roce_gid_table is used.
2367 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2368 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2370 struct ib_ucontext *(*alloc_ucontext)(struct ib_device *device,
2371 struct ib_udata *udata);
2372 int (*dealloc_ucontext)(struct ib_ucontext *context);
2373 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2374 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2375 struct ib_pd *(*alloc_pd)(struct ib_device *device,
2376 struct ib_ucontext *context,
2377 struct ib_udata *udata);
2378 int (*dealloc_pd)(struct ib_pd *pd);
2379 struct ib_ah *(*create_ah)(struct ib_pd *pd,
2380 struct rdma_ah_attr *ah_attr, u32 flags,
2381 struct ib_udata *udata);
2382 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2383 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2384 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2385 struct ib_srq *(*create_srq)(struct ib_pd *pd,
2386 struct ib_srq_init_attr *srq_init_attr,
2387 struct ib_udata *udata);
2388 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2389 enum ib_srq_attr_mask srq_attr_mask,
2390 struct ib_udata *udata);
2391 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2392 int (*destroy_srq)(struct ib_srq *srq);
2393 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2394 struct ib_qp_init_attr *qp_init_attr,
2395 struct ib_udata *udata);
2396 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2397 int qp_attr_mask, struct ib_udata *udata);
2398 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2399 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2400 int (*destroy_qp)(struct ib_qp *qp);
2401 struct ib_cq *(*create_cq)(struct ib_device *device,
2402 const struct ib_cq_init_attr *attr,
2403 struct ib_ucontext *context,
2404 struct ib_udata *udata);
2405 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2406 int (*destroy_cq)(struct ib_cq *cq);
2407 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2408 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2409 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2410 u64 virt_addr, int mr_access_flags,
2411 struct ib_udata *udata);
2412 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2413 u64 virt_addr, int mr_access_flags,
2414 struct ib_pd *pd, struct ib_udata *udata);
2415 int (*dereg_mr)(struct ib_mr *mr);
2416 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2418 int (*advise_mr)(struct ib_pd *pd,
2419 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2420 struct ib_sge *sg_list, u32 num_sge,
2421 struct uverbs_attr_bundle *attrs);
2422 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2423 unsigned int *sg_offset);
2424 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2425 struct ib_mr_status *mr_status);
2426 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2427 struct ib_udata *udata);
2428 int (*dealloc_mw)(struct ib_mw *mw);
2429 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2430 struct ib_fmr_attr *fmr_attr);
2431 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2433 int (*unmap_fmr)(struct list_head *fmr_list);
2434 int (*dealloc_fmr)(struct ib_fmr *fmr);
2435 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2436 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2437 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2438 struct ib_ucontext *ucontext,
2439 struct ib_udata *udata);
2440 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2441 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2442 struct ib_flow_attr *flow_attr,
2443 int domain, struct ib_udata *udata);
2444 int (*destroy_flow)(struct ib_flow *flow_id);
2445 struct ib_flow_action *(*create_flow_action_esp)(
2446 struct ib_device *device,
2447 const struct ib_flow_action_attrs_esp *attr,
2448 struct uverbs_attr_bundle *attrs);
2449 int (*destroy_flow_action)(struct ib_flow_action *action);
2450 int (*modify_flow_action_esp)(
2451 struct ib_flow_action *action,
2452 const struct ib_flow_action_attrs_esp *attr,
2453 struct uverbs_attr_bundle *attrs);
2454 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2456 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2457 struct ifla_vf_info *ivf);
2458 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2459 struct ifla_vf_stats *stats);
2460 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2462 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2463 struct ib_wq_init_attr *init_attr,
2464 struct ib_udata *udata);
2465 int (*destroy_wq)(struct ib_wq *wq);
2466 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2467 u32 wq_attr_mask, struct ib_udata *udata);
2468 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2469 struct ib_device *device,
2470 struct ib_rwq_ind_table_init_attr *init_attr,
2471 struct ib_udata *udata);
2472 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2473 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2474 struct ib_ucontext *context,
2475 struct ib_dm_alloc_attr *attr,
2476 struct uverbs_attr_bundle *attrs);
2477 int (*dealloc_dm)(struct ib_dm *dm);
2478 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2479 struct ib_dm_mr_attr *attr,
2480 struct uverbs_attr_bundle *attrs);
2481 struct ib_counters *(*create_counters)(
2482 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2483 int (*destroy_counters)(struct ib_counters *counters);
2484 int (*read_counters)(struct ib_counters *counters,
2485 struct ib_counters_read_attr *counters_read_attr,
2486 struct uverbs_attr_bundle *attrs);
2488 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2489 * driver initialized data. The struct is kfree()'ed by the sysfs
2490 * core when the device is removed. A lifespan of -1 in the return
2491 * struct tells the core to set a default lifespan.
2493 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2496 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2497 * @index - The index in the value array we wish to have updated, or
2498 * num_counters if we want all stats updated
2500 * < 0 - Error, no counters updated
2501 * index - Updated the single counter pointed to by index
2502 * num_counters - Updated all counters (will reset the timestamp
2503 * and prevent further calls for lifespan milliseconds)
2504 * Drivers are allowed to update all counters in leiu of just the
2505 * one given in index at their option
2507 int (*get_hw_stats)(struct ib_device *device,
2508 struct rdma_hw_stats *stats, u8 port, int index);
2512 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2513 struct device *dma_device;
2514 struct ib_device_ops ops;
2515 char name[IB_DEVICE_NAME_MAX];
2517 struct list_head event_handler_list;
2518 spinlock_t event_handler_lock;
2520 rwlock_t client_data_lock;
2521 struct list_head core_list;
2522 /* Access to the client_data_list is protected by the client_data_lock
2523 * rwlock and the lists_rwsem read-write semaphore
2525 struct list_head client_data_list;
2527 struct ib_cache cache;
2529 * port_immutable is indexed by port number
2531 struct ib_port_immutable *port_immutable;
2533 int num_comp_vectors;
2535 struct ib_port_pkey_list *port_pkey_list;
2537 struct iw_cm_verbs *iwcm;
2539 struct module *owner;
2541 /* First group for device attributes,
2542 * Second group for driver provided attributes (optional).
2543 * It is NULL terminated array.
2545 const struct attribute_group *groups[3];
2547 struct kobject *ports_kobj;
2548 struct list_head port_list;
2551 IB_DEV_UNINITIALIZED,
2557 u64 uverbs_cmd_mask;
2558 u64 uverbs_ex_cmd_mask;
2560 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2566 struct ib_device_attr attrs;
2567 struct attribute_group *hw_stats_ag;
2568 struct rdma_hw_stats *hw_stats;
2570 #ifdef CONFIG_CGROUP_RDMA
2571 struct rdmacg_device cg_device;
2576 * Implementation details of the RDMA core, don't use in drivers
2578 struct rdma_restrack_root res;
2580 const struct uapi_definition *driver_def;
2581 enum rdma_driver_id driver_id;
2583 * Provides synchronization between device unregistration and netlink
2584 * commands on a device. To be used only by core.
2586 refcount_t refcount;
2587 struct completion unreg_completion;
2592 void (*add) (struct ib_device *);
2593 void (*remove)(struct ib_device *, void *client_data);
2595 /* Returns the net_dev belonging to this ib_client and matching the
2597 * @dev: An RDMA device that the net_dev use for communication.
2598 * @port: A physical port number on the RDMA device.
2599 * @pkey: P_Key that the net_dev uses if applicable.
2600 * @gid: A GID that the net_dev uses to communicate.
2601 * @addr: An IP address the net_dev is configured with.
2602 * @client_data: The device's client data set by ib_set_client_data().
2604 * An ib_client that implements a net_dev on top of RDMA devices
2605 * (such as IP over IB) should implement this callback, allowing the
2606 * rdma_cm module to find the right net_dev for a given request.
2608 * The caller is responsible for calling dev_put on the returned
2610 struct net_device *(*get_net_dev_by_params)(
2611 struct ib_device *dev,
2614 const union ib_gid *gid,
2615 const struct sockaddr *addr,
2617 struct list_head list;
2620 struct ib_device *ib_alloc_device(size_t size);
2621 void ib_dealloc_device(struct ib_device *device);
2623 void ib_get_device_fw_str(struct ib_device *device, char *str);
2625 int ib_register_device(struct ib_device *device, const char *name,
2626 int (*port_callback)(struct ib_device *, u8,
2628 void ib_unregister_device(struct ib_device *device);
2630 int ib_register_client (struct ib_client *client);
2631 void ib_unregister_client(struct ib_client *client);
2633 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2634 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2636 void ib_set_device_ops(struct ib_device *device,
2637 const struct ib_device_ops *ops);
2639 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2640 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2641 unsigned long pfn, unsigned long size, pgprot_t prot);
2642 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2643 struct vm_area_struct *vma, struct page *page,
2644 unsigned long size);
2646 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2647 struct vm_area_struct *vma,
2648 unsigned long pfn, unsigned long size,
2653 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2654 struct vm_area_struct *vma, struct page *page,
2661 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2663 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2666 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2668 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2671 static inline bool ib_is_buffer_cleared(const void __user *p,
2677 if (len > USHRT_MAX)
2680 buf = memdup_user(p, len);
2684 ret = !memchr_inv(buf, 0, len);
2689 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2693 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2697 * ib_is_destroy_retryable - Check whether the uobject destruction
2699 * @ret: The initial destruction return code
2700 * @why: remove reason
2701 * @uobj: The uobject that is destroyed
2703 * This function is a helper function that IB layer and low-level drivers
2704 * can use to consider whether the destruction of the given uobject is
2706 * It checks the original return code, if it wasn't success the destruction
2707 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2708 * the remove reason. (i.e. why).
2709 * Must be called with the object locked for destroy.
2711 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2712 struct ib_uobject *uobj)
2714 return ret && (why == RDMA_REMOVE_DESTROY ||
2715 uobj->context->cleanup_retryable);
2719 * ib_destroy_usecnt - Called during destruction to check the usecnt
2720 * @usecnt: The usecnt atomic
2721 * @why: remove reason
2722 * @uobj: The uobject that is destroyed
2724 * Non-zero usecnts will block destruction unless destruction was triggered by
2725 * a ucontext cleanup.
2727 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2728 enum rdma_remove_reason why,
2729 struct ib_uobject *uobj)
2731 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2737 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2738 * contains all required attributes and no attributes not allowed for
2739 * the given QP state transition.
2740 * @cur_state: Current QP state
2741 * @next_state: Next QP state
2743 * @mask: Mask of supplied QP attributes
2745 * This function is a helper function that a low-level driver's
2746 * modify_qp method can use to validate the consumer's input. It
2747 * checks that cur_state and next_state are valid QP states, that a
2748 * transition from cur_state to next_state is allowed by the IB spec,
2749 * and that the attribute mask supplied is allowed for the transition.
2751 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2752 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2754 void ib_register_event_handler(struct ib_event_handler *event_handler);
2755 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2756 void ib_dispatch_event(struct ib_event *event);
2758 int ib_query_port(struct ib_device *device,
2759 u8 port_num, struct ib_port_attr *port_attr);
2761 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2765 * rdma_cap_ib_switch - Check if the device is IB switch
2766 * @device: Device to check
2768 * Device driver is responsible for setting is_switch bit on
2769 * in ib_device structure at init time.
2771 * Return: true if the device is IB switch.
2773 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2775 return device->is_switch;
2779 * rdma_start_port - Return the first valid port number for the device
2782 * @device: Device to be checked
2784 * Return start port number
2786 static inline u8 rdma_start_port(const struct ib_device *device)
2788 return rdma_cap_ib_switch(device) ? 0 : 1;
2792 * rdma_end_port - Return the last valid port number for the device
2795 * @device: Device to be checked
2797 * Return last port number
2799 static inline u8 rdma_end_port(const struct ib_device *device)
2801 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2804 static inline int rdma_is_port_valid(const struct ib_device *device,
2807 return (port >= rdma_start_port(device) &&
2808 port <= rdma_end_port(device));
2811 static inline bool rdma_is_grh_required(const struct ib_device *device,
2814 return device->port_immutable[port_num].core_cap_flags &
2815 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2818 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2823 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2825 return device->port_immutable[port_num].core_cap_flags &
2826 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2829 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2831 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2834 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2836 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2839 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2841 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2844 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2846 return rdma_protocol_ib(device, port_num) ||
2847 rdma_protocol_roce(device, port_num);
2850 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2852 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2855 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2857 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2861 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2862 * Management Datagrams.
2863 * @device: Device to check
2864 * @port_num: Port number to check
2866 * Management Datagrams (MAD) are a required part of the InfiniBand
2867 * specification and are supported on all InfiniBand devices. A slightly
2868 * extended version are also supported on OPA interfaces.
2870 * Return: true if the port supports sending/receiving of MAD packets.
2872 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2874 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2878 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2879 * Management Datagrams.
2880 * @device: Device to check
2881 * @port_num: Port number to check
2883 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2884 * datagrams with their own versions. These OPA MADs share many but not all of
2885 * the characteristics of InfiniBand MADs.
2887 * OPA MADs differ in the following ways:
2889 * 1) MADs are variable size up to 2K
2890 * IBTA defined MADs remain fixed at 256 bytes
2891 * 2) OPA SMPs must carry valid PKeys
2892 * 3) OPA SMP packets are a different format
2894 * Return: true if the port supports OPA MAD packet formats.
2896 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2898 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2899 == RDMA_CORE_CAP_OPA_MAD;
2903 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2904 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2905 * @device: Device to check
2906 * @port_num: Port number to check
2908 * Each InfiniBand node is required to provide a Subnet Management Agent
2909 * that the subnet manager can access. Prior to the fabric being fully
2910 * configured by the subnet manager, the SMA is accessed via a well known
2911 * interface called the Subnet Management Interface (SMI). This interface
2912 * uses directed route packets to communicate with the SM to get around the
2913 * chicken and egg problem of the SM needing to know what's on the fabric
2914 * in order to configure the fabric, and needing to configure the fabric in
2915 * order to send packets to the devices on the fabric. These directed
2916 * route packets do not need the fabric fully configured in order to reach
2917 * their destination. The SMI is the only method allowed to send
2918 * directed route packets on an InfiniBand fabric.
2920 * Return: true if the port provides an SMI.
2922 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2924 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2928 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2929 * Communication Manager.
2930 * @device: Device to check
2931 * @port_num: Port number to check
2933 * The InfiniBand Communication Manager is one of many pre-defined General
2934 * Service Agents (GSA) that are accessed via the General Service
2935 * Interface (GSI). It's role is to facilitate establishment of connections
2936 * between nodes as well as other management related tasks for established
2939 * Return: true if the port supports an IB CM (this does not guarantee that
2940 * a CM is actually running however).
2942 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2944 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2948 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2949 * Communication Manager.
2950 * @device: Device to check
2951 * @port_num: Port number to check
2953 * Similar to above, but specific to iWARP connections which have a different
2954 * managment protocol than InfiniBand.
2956 * Return: true if the port supports an iWARP CM (this does not guarantee that
2957 * a CM is actually running however).
2959 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2961 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2965 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2966 * Subnet Administration.
2967 * @device: Device to check
2968 * @port_num: Port number to check
2970 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2971 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2972 * fabrics, devices should resolve routes to other hosts by contacting the
2973 * SA to query the proper route.
2975 * Return: true if the port should act as a client to the fabric Subnet
2976 * Administration interface. This does not imply that the SA service is
2979 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2981 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2985 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2987 * @device: Device to check
2988 * @port_num: Port number to check
2990 * InfiniBand multicast registration is more complex than normal IPv4 or
2991 * IPv6 multicast registration. Each Host Channel Adapter must register
2992 * with the Subnet Manager when it wishes to join a multicast group. It
2993 * should do so only once regardless of how many queue pairs it subscribes
2994 * to this group. And it should leave the group only after all queue pairs
2995 * attached to the group have been detached.
2997 * Return: true if the port must undertake the additional adminstrative
2998 * overhead of registering/unregistering with the SM and tracking of the
2999 * total number of queue pairs attached to the multicast group.
3001 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3003 return rdma_cap_ib_sa(device, port_num);
3007 * rdma_cap_af_ib - Check if the port of device has the capability
3008 * Native Infiniband Address.
3009 * @device: Device to check
3010 * @port_num: Port number to check
3012 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3013 * GID. RoCE uses a different mechanism, but still generates a GID via
3014 * a prescribed mechanism and port specific data.
3016 * Return: true if the port uses a GID address to identify devices on the
3019 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3021 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3025 * rdma_cap_eth_ah - Check if the port of device has the capability
3026 * Ethernet Address Handle.
3027 * @device: Device to check
3028 * @port_num: Port number to check
3030 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3031 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3032 * port. Normally, packet headers are generated by the sending host
3033 * adapter, but when sending connectionless datagrams, we must manually
3034 * inject the proper headers for the fabric we are communicating over.
3036 * Return: true if we are running as a RoCE port and must force the
3037 * addition of a Global Route Header built from our Ethernet Address
3038 * Handle into our header list for connectionless packets.
3040 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3042 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3046 * rdma_cap_opa_ah - Check if the port of device supports
3047 * OPA Address handles
3048 * @device: Device to check
3049 * @port_num: Port number to check
3051 * Return: true if we are running on an OPA device which supports
3052 * the extended OPA addressing.
3054 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3056 return (device->port_immutable[port_num].core_cap_flags &
3057 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3061 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3064 * @port_num: Port number
3066 * This MAD size includes the MAD headers and MAD payload. No other headers
3069 * Return the max MAD size required by the Port. Will return 0 if the port
3070 * does not support MADs
3072 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3074 return device->port_immutable[port_num].max_mad_size;
3078 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3079 * @device: Device to check
3080 * @port_num: Port number to check
3082 * RoCE GID table mechanism manages the various GIDs for a device.
3084 * NOTE: if allocating the port's GID table has failed, this call will still
3085 * return true, but any RoCE GID table API will fail.
3087 * Return: true if the port uses RoCE GID table mechanism in order to manage
3090 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3093 return rdma_protocol_roce(device, port_num) &&
3094 device->ops.add_gid && device->ops.del_gid;
3098 * Check if the device supports READ W/ INVALIDATE.
3100 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3103 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3104 * has support for it yet.
3106 return rdma_protocol_iwarp(dev, port_num);
3109 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3111 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3112 struct ifla_vf_info *info);
3113 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3114 struct ifla_vf_stats *stats);
3115 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3118 int ib_query_pkey(struct ib_device *device,
3119 u8 port_num, u16 index, u16 *pkey);
3121 int ib_modify_device(struct ib_device *device,
3122 int device_modify_mask,
3123 struct ib_device_modify *device_modify);
3125 int ib_modify_port(struct ib_device *device,
3126 u8 port_num, int port_modify_mask,
3127 struct ib_port_modify *port_modify);
3129 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3130 u8 *port_num, u16 *index);
3132 int ib_find_pkey(struct ib_device *device,
3133 u8 port_num, u16 pkey, u16 *index);
3137 * Create a memory registration for all memory in the system and place
3138 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3139 * ULPs to avoid the overhead of dynamic MRs.
3141 * This flag is generally considered unsafe and must only be used in
3142 * extremly trusted environments. Every use of it will log a warning
3143 * in the kernel log.
3145 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3148 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3149 const char *caller);
3150 #define ib_alloc_pd(device, flags) \
3151 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3152 void ib_dealloc_pd(struct ib_pd *pd);
3154 enum rdma_create_ah_flags {
3155 /* In a sleepable context */
3156 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3160 * rdma_create_ah - Creates an address handle for the given address vector.
3161 * @pd: The protection domain associated with the address handle.
3162 * @ah_attr: The attributes of the address vector.
3163 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3165 * The address handle is used to reference a local or global destination
3166 * in all UD QP post sends.
3168 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3172 * rdma_create_user_ah - Creates an address handle for the given address vector.
3173 * It resolves destination mac address for ah attribute of RoCE type.
3174 * @pd: The protection domain associated with the address handle.
3175 * @ah_attr: The attributes of the address vector.
3176 * @udata: pointer to user's input output buffer information need by
3179 * It returns 0 on success and returns appropriate error code on error.
3180 * The address handle is used to reference a local or global destination
3181 * in all UD QP post sends.
3183 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3184 struct rdma_ah_attr *ah_attr,
3185 struct ib_udata *udata);
3187 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3189 * @hdr: the L3 header to parse
3190 * @net_type: type of header to parse
3191 * @sgid: place to store source gid
3192 * @dgid: place to store destination gid
3194 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3195 enum rdma_network_type net_type,
3196 union ib_gid *sgid, union ib_gid *dgid);
3199 * ib_get_rdma_header_version - Get the header version
3200 * @hdr: the L3 header to parse
3202 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3205 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3207 * @device: Device on which the received message arrived.
3208 * @port_num: Port on which the received message arrived.
3209 * @wc: Work completion associated with the received message.
3210 * @grh: References the received global route header. This parameter is
3211 * ignored unless the work completion indicates that the GRH is valid.
3212 * @ah_attr: Returned attributes that can be used when creating an address
3213 * handle for replying to the message.
3214 * When ib_init_ah_attr_from_wc() returns success,
3215 * (a) for IB link layer it optionally contains a reference to SGID attribute
3216 * when GRH is present for IB link layer.
3217 * (b) for RoCE link layer it contains a reference to SGID attribute.
3218 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3219 * attributes which are initialized using ib_init_ah_attr_from_wc().
3222 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3223 const struct ib_wc *wc, const struct ib_grh *grh,
3224 struct rdma_ah_attr *ah_attr);
3227 * ib_create_ah_from_wc - Creates an address handle associated with the
3228 * sender of the specified work completion.
3229 * @pd: The protection domain associated with the address handle.
3230 * @wc: Work completion information associated with a received message.
3231 * @grh: References the received global route header. This parameter is
3232 * ignored unless the work completion indicates that the GRH is valid.
3233 * @port_num: The outbound port number to associate with the address.
3235 * The address handle is used to reference a local or global destination
3236 * in all UD QP post sends.
3238 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3239 const struct ib_grh *grh, u8 port_num);
3242 * rdma_modify_ah - Modifies the address vector associated with an address
3244 * @ah: The address handle to modify.
3245 * @ah_attr: The new address vector attributes to associate with the
3248 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3251 * rdma_query_ah - Queries the address vector associated with an address
3253 * @ah: The address handle to query.
3254 * @ah_attr: The address vector attributes associated with the address
3257 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3259 enum rdma_destroy_ah_flags {
3260 /* In a sleepable context */
3261 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3265 * rdma_destroy_ah - Destroys an address handle.
3266 * @ah: The address handle to destroy.
3267 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3269 int rdma_destroy_ah(struct ib_ah *ah, u32 flags);
3272 * ib_create_srq - Creates a SRQ associated with the specified protection
3274 * @pd: The protection domain associated with the SRQ.
3275 * @srq_init_attr: A list of initial attributes required to create the
3276 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3277 * the actual capabilities of the created SRQ.
3279 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3280 * requested size of the SRQ, and set to the actual values allocated
3281 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3282 * will always be at least as large as the requested values.
3284 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3285 struct ib_srq_init_attr *srq_init_attr);
3288 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3289 * @srq: The SRQ to modify.
3290 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3291 * the current values of selected SRQ attributes are returned.
3292 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3293 * are being modified.
3295 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3296 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3297 * the number of receives queued drops below the limit.
3299 int ib_modify_srq(struct ib_srq *srq,
3300 struct ib_srq_attr *srq_attr,
3301 enum ib_srq_attr_mask srq_attr_mask);
3304 * ib_query_srq - Returns the attribute list and current values for the
3306 * @srq: The SRQ to query.
3307 * @srq_attr: The attributes of the specified SRQ.
3309 int ib_query_srq(struct ib_srq *srq,
3310 struct ib_srq_attr *srq_attr);
3313 * ib_destroy_srq - Destroys the specified SRQ.
3314 * @srq: The SRQ to destroy.
3316 int ib_destroy_srq(struct ib_srq *srq);
3319 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3320 * @srq: The SRQ to post the work request on.
3321 * @recv_wr: A list of work requests to post on the receive queue.
3322 * @bad_recv_wr: On an immediate failure, this parameter will reference
3323 * the work request that failed to be posted on the QP.
3325 static inline int ib_post_srq_recv(struct ib_srq *srq,
3326 const struct ib_recv_wr *recv_wr,
3327 const struct ib_recv_wr **bad_recv_wr)
3329 const struct ib_recv_wr *dummy;
3331 return srq->device->ops.post_srq_recv(srq, recv_wr,
3332 bad_recv_wr ? : &dummy);
3336 * ib_create_qp - Creates a QP associated with the specified protection
3338 * @pd: The protection domain associated with the QP.
3339 * @qp_init_attr: A list of initial attributes required to create the
3340 * QP. If QP creation succeeds, then the attributes are updated to
3341 * the actual capabilities of the created QP.
3343 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3344 struct ib_qp_init_attr *qp_init_attr);
3347 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3348 * @qp: The QP to modify.
3349 * @attr: On input, specifies the QP attributes to modify. On output,
3350 * the current values of selected QP attributes are returned.
3351 * @attr_mask: A bit-mask used to specify which attributes of the QP
3352 * are being modified.
3353 * @udata: pointer to user's input output buffer information
3354 * are being modified.
3355 * It returns 0 on success and returns appropriate error code on error.
3357 int ib_modify_qp_with_udata(struct ib_qp *qp,
3358 struct ib_qp_attr *attr,
3360 struct ib_udata *udata);
3363 * ib_modify_qp - Modifies the attributes for the specified QP and then
3364 * transitions the QP to the given state.
3365 * @qp: The QP to modify.
3366 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3367 * the current values of selected QP attributes are returned.
3368 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3369 * are being modified.
3371 int ib_modify_qp(struct ib_qp *qp,
3372 struct ib_qp_attr *qp_attr,
3376 * ib_query_qp - Returns the attribute list and current values for the
3378 * @qp: The QP to query.
3379 * @qp_attr: The attributes of the specified QP.
3380 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3381 * @qp_init_attr: Additional attributes of the selected QP.
3383 * The qp_attr_mask may be used to limit the query to gathering only the
3384 * selected attributes.
3386 int ib_query_qp(struct ib_qp *qp,
3387 struct ib_qp_attr *qp_attr,
3389 struct ib_qp_init_attr *qp_init_attr);
3392 * ib_destroy_qp - Destroys the specified QP.
3393 * @qp: The QP to destroy.
3395 int ib_destroy_qp(struct ib_qp *qp);
3398 * ib_open_qp - Obtain a reference to an existing sharable QP.
3399 * @xrcd - XRC domain
3400 * @qp_open_attr: Attributes identifying the QP to open.
3402 * Returns a reference to a sharable QP.
3404 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3405 struct ib_qp_open_attr *qp_open_attr);
3408 * ib_close_qp - Release an external reference to a QP.
3409 * @qp: The QP handle to release
3411 * The opened QP handle is released by the caller. The underlying
3412 * shared QP is not destroyed until all internal references are released.
3414 int ib_close_qp(struct ib_qp *qp);
3417 * ib_post_send - Posts a list of work requests to the send queue of
3419 * @qp: The QP to post the work request on.
3420 * @send_wr: A list of work requests to post on the send queue.
3421 * @bad_send_wr: On an immediate failure, this parameter will reference
3422 * the work request that failed to be posted on the QP.
3424 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3425 * error is returned, the QP state shall not be affected,
3426 * ib_post_send() will return an immediate error after queueing any
3427 * earlier work requests in the list.
3429 static inline int ib_post_send(struct ib_qp *qp,
3430 const struct ib_send_wr *send_wr,
3431 const struct ib_send_wr **bad_send_wr)
3433 const struct ib_send_wr *dummy;
3435 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3439 * ib_post_recv - Posts a list of work requests to the receive queue of
3441 * @qp: The QP to post the work request on.
3442 * @recv_wr: A list of work requests to post on the receive queue.
3443 * @bad_recv_wr: On an immediate failure, this parameter will reference
3444 * the work request that failed to be posted on the QP.
3446 static inline int ib_post_recv(struct ib_qp *qp,
3447 const struct ib_recv_wr *recv_wr,
3448 const struct ib_recv_wr **bad_recv_wr)
3450 const struct ib_recv_wr *dummy;
3452 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3455 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3456 int nr_cqe, int comp_vector,
3457 enum ib_poll_context poll_ctx, const char *caller);
3458 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3459 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3461 void ib_free_cq(struct ib_cq *cq);
3462 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3465 * ib_create_cq - Creates a CQ on the specified device.
3466 * @device: The device on which to create the CQ.
3467 * @comp_handler: A user-specified callback that is invoked when a
3468 * completion event occurs on the CQ.
3469 * @event_handler: A user-specified callback that is invoked when an
3470 * asynchronous event not associated with a completion occurs on the CQ.
3471 * @cq_context: Context associated with the CQ returned to the user via
3472 * the associated completion and event handlers.
3473 * @cq_attr: The attributes the CQ should be created upon.
3475 * Users can examine the cq structure to determine the actual CQ size.
3477 struct ib_cq *__ib_create_cq(struct ib_device *device,
3478 ib_comp_handler comp_handler,
3479 void (*event_handler)(struct ib_event *, void *),
3481 const struct ib_cq_init_attr *cq_attr,
3482 const char *caller);
3483 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3484 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3487 * ib_resize_cq - Modifies the capacity of the CQ.
3488 * @cq: The CQ to resize.
3489 * @cqe: The minimum size of the CQ.
3491 * Users can examine the cq structure to determine the actual CQ size.
3493 int ib_resize_cq(struct ib_cq *cq, int cqe);
3496 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3497 * @cq: The CQ to modify.
3498 * @cq_count: number of CQEs that will trigger an event
3499 * @cq_period: max period of time in usec before triggering an event
3502 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3505 * ib_destroy_cq - Destroys the specified CQ.
3506 * @cq: The CQ to destroy.
3508 int ib_destroy_cq(struct ib_cq *cq);
3511 * ib_poll_cq - poll a CQ for completion(s)
3512 * @cq:the CQ being polled
3513 * @num_entries:maximum number of completions to return
3514 * @wc:array of at least @num_entries &struct ib_wc where completions
3517 * Poll a CQ for (possibly multiple) completions. If the return value
3518 * is < 0, an error occurred. If the return value is >= 0, it is the
3519 * number of completions returned. If the return value is
3520 * non-negative and < num_entries, then the CQ was emptied.
3522 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3525 return cq->device->ops.poll_cq(cq, num_entries, wc);
3529 * ib_req_notify_cq - Request completion notification on a CQ.
3530 * @cq: The CQ to generate an event for.
3532 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3533 * to request an event on the next solicited event or next work
3534 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3535 * may also be |ed in to request a hint about missed events, as
3539 * < 0 means an error occurred while requesting notification
3540 * == 0 means notification was requested successfully, and if
3541 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3542 * were missed and it is safe to wait for another event. In
3543 * this case is it guaranteed that any work completions added
3544 * to the CQ since the last CQ poll will trigger a completion
3545 * notification event.
3546 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3547 * in. It means that the consumer must poll the CQ again to
3548 * make sure it is empty to avoid missing an event because of a
3549 * race between requesting notification and an entry being
3550 * added to the CQ. This return value means it is possible
3551 * (but not guaranteed) that a work completion has been added
3552 * to the CQ since the last poll without triggering a
3553 * completion notification event.
3555 static inline int ib_req_notify_cq(struct ib_cq *cq,
3556 enum ib_cq_notify_flags flags)
3558 return cq->device->ops.req_notify_cq(cq, flags);
3562 * ib_req_ncomp_notif - Request completion notification when there are
3563 * at least the specified number of unreaped completions on the CQ.
3564 * @cq: The CQ to generate an event for.
3565 * @wc_cnt: The number of unreaped completions that should be on the
3566 * CQ before an event is generated.
3568 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3570 return cq->device->ops.req_ncomp_notif ?
3571 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3576 * ib_dma_mapping_error - check a DMA addr for error
3577 * @dev: The device for which the dma_addr was created
3578 * @dma_addr: The DMA address to check
3580 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3582 return dma_mapping_error(dev->dma_device, dma_addr);
3586 * ib_dma_map_single - Map a kernel virtual address to DMA address
3587 * @dev: The device for which the dma_addr is to be created
3588 * @cpu_addr: The kernel virtual address
3589 * @size: The size of the region in bytes
3590 * @direction: The direction of the DMA
3592 static inline u64 ib_dma_map_single(struct ib_device *dev,
3593 void *cpu_addr, size_t size,
3594 enum dma_data_direction direction)
3596 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3600 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3601 * @dev: The device for which the DMA address was created
3602 * @addr: The DMA address
3603 * @size: The size of the region in bytes
3604 * @direction: The direction of the DMA
3606 static inline void ib_dma_unmap_single(struct ib_device *dev,
3607 u64 addr, size_t size,
3608 enum dma_data_direction direction)
3610 dma_unmap_single(dev->dma_device, addr, size, direction);
3614 * ib_dma_map_page - Map a physical page to DMA address
3615 * @dev: The device for which the dma_addr is to be created
3616 * @page: The page to be mapped
3617 * @offset: The offset within the page
3618 * @size: The size of the region in bytes
3619 * @direction: The direction of the DMA
3621 static inline u64 ib_dma_map_page(struct ib_device *dev,
3623 unsigned long offset,
3625 enum dma_data_direction direction)
3627 return dma_map_page(dev->dma_device, page, offset, size, direction);
3631 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3632 * @dev: The device for which the DMA address was created
3633 * @addr: The DMA address
3634 * @size: The size of the region in bytes
3635 * @direction: The direction of the DMA
3637 static inline void ib_dma_unmap_page(struct ib_device *dev,
3638 u64 addr, size_t size,
3639 enum dma_data_direction direction)
3641 dma_unmap_page(dev->dma_device, addr, size, direction);
3645 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3646 * @dev: The device for which the DMA addresses are to be created
3647 * @sg: The array of scatter/gather entries
3648 * @nents: The number of scatter/gather entries
3649 * @direction: The direction of the DMA
3651 static inline int ib_dma_map_sg(struct ib_device *dev,
3652 struct scatterlist *sg, int nents,
3653 enum dma_data_direction direction)
3655 return dma_map_sg(dev->dma_device, sg, nents, direction);
3659 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3660 * @dev: The device for which the DMA addresses were created
3661 * @sg: The array of scatter/gather entries
3662 * @nents: The number of scatter/gather entries
3663 * @direction: The direction of the DMA
3665 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3666 struct scatterlist *sg, int nents,
3667 enum dma_data_direction direction)
3669 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3672 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3673 struct scatterlist *sg, int nents,
3674 enum dma_data_direction direction,
3675 unsigned long dma_attrs)
3677 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3681 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3682 struct scatterlist *sg, int nents,
3683 enum dma_data_direction direction,
3684 unsigned long dma_attrs)
3686 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3689 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3690 * @dev: The device for which the DMA addresses were created
3691 * @sg: The scatter/gather entry
3693 * Note: this function is obsolete. To do: change all occurrences of
3694 * ib_sg_dma_address() into sg_dma_address().
3696 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3697 struct scatterlist *sg)
3699 return sg_dma_address(sg);
3703 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3704 * @dev: The device for which the DMA addresses were created
3705 * @sg: The scatter/gather entry
3707 * Note: this function is obsolete. To do: change all occurrences of
3708 * ib_sg_dma_len() into sg_dma_len().
3710 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3711 struct scatterlist *sg)
3713 return sg_dma_len(sg);
3717 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3718 * @dev: The device for which the DMA address was created
3719 * @addr: The DMA address
3720 * @size: The size of the region in bytes
3721 * @dir: The direction of the DMA
3723 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3726 enum dma_data_direction dir)
3728 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3732 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3733 * @dev: The device for which the DMA address was created
3734 * @addr: The DMA address
3735 * @size: The size of the region in bytes
3736 * @dir: The direction of the DMA
3738 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3741 enum dma_data_direction dir)
3743 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3747 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3748 * @dev: The device for which the DMA address is requested
3749 * @size: The size of the region to allocate in bytes
3750 * @dma_handle: A pointer for returning the DMA address of the region
3751 * @flag: memory allocator flags
3753 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3755 dma_addr_t *dma_handle,
3758 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3762 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3763 * @dev: The device for which the DMA addresses were allocated
3764 * @size: The size of the region
3765 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3766 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3768 static inline void ib_dma_free_coherent(struct ib_device *dev,
3769 size_t size, void *cpu_addr,
3770 dma_addr_t dma_handle)
3772 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3776 * ib_dereg_mr - Deregisters a memory region and removes it from the
3777 * HCA translation table.
3778 * @mr: The memory region to deregister.
3780 * This function can fail, if the memory region has memory windows bound to it.
3782 int ib_dereg_mr(struct ib_mr *mr);
3784 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3785 enum ib_mr_type mr_type,
3789 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3791 * @mr - struct ib_mr pointer to be updated.
3792 * @newkey - new key to be used.
3794 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3796 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3797 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3801 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3802 * for calculating a new rkey for type 2 memory windows.
3803 * @rkey - the rkey to increment.
3805 static inline u32 ib_inc_rkey(u32 rkey)
3807 const u32 mask = 0x000000ff;
3808 return ((rkey + 1) & mask) | (rkey & ~mask);
3812 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3813 * @pd: The protection domain associated with the unmapped region.
3814 * @mr_access_flags: Specifies the memory access rights.
3815 * @fmr_attr: Attributes of the unmapped region.
3817 * A fast memory region must be mapped before it can be used as part of
3820 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3821 int mr_access_flags,
3822 struct ib_fmr_attr *fmr_attr);
3825 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3826 * @fmr: The fast memory region to associate with the pages.
3827 * @page_list: An array of physical pages to map to the fast memory region.
3828 * @list_len: The number of pages in page_list.
3829 * @iova: The I/O virtual address to use with the mapped region.
3831 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3832 u64 *page_list, int list_len,
3835 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
3839 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3840 * @fmr_list: A linked list of fast memory regions to unmap.
3842 int ib_unmap_fmr(struct list_head *fmr_list);
3845 * ib_dealloc_fmr - Deallocates a fast memory region.
3846 * @fmr: The fast memory region to deallocate.
3848 int ib_dealloc_fmr(struct ib_fmr *fmr);
3851 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3852 * @qp: QP to attach to the multicast group. The QP must be type
3854 * @gid: Multicast group GID.
3855 * @lid: Multicast group LID in host byte order.
3857 * In order to send and receive multicast packets, subnet
3858 * administration must have created the multicast group and configured
3859 * the fabric appropriately. The port associated with the specified
3860 * QP must also be a member of the multicast group.
3862 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3865 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3866 * @qp: QP to detach from the multicast group.
3867 * @gid: Multicast group GID.
3868 * @lid: Multicast group LID in host byte order.
3870 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3873 * ib_alloc_xrcd - Allocates an XRC domain.
3874 * @device: The device on which to allocate the XRC domain.
3875 * @caller: Module name for kernel consumers
3877 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3878 #define ib_alloc_xrcd(device) \
3879 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3882 * ib_dealloc_xrcd - Deallocates an XRC domain.
3883 * @xrcd: The XRC domain to deallocate.
3885 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3887 static inline int ib_check_mr_access(int flags)
3890 * Local write permission is required if remote write or
3891 * remote atomic permission is also requested.
3893 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3894 !(flags & IB_ACCESS_LOCAL_WRITE))
3900 static inline bool ib_access_writable(int access_flags)
3903 * We have writable memory backing the MR if any of the following
3904 * access flags are set. "Local write" and "remote write" obviously
3905 * require write access. "Remote atomic" can do things like fetch and
3906 * add, which will modify memory, and "MW bind" can change permissions
3907 * by binding a window.
3909 return access_flags &
3910 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3911 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3915 * ib_check_mr_status: lightweight check of MR status.
3916 * This routine may provide status checks on a selected
3917 * ib_mr. first use is for signature status check.
3919 * @mr: A memory region.
3920 * @check_mask: Bitmask of which checks to perform from
3921 * ib_mr_status_check enumeration.
3922 * @mr_status: The container of relevant status checks.
3923 * failed checks will be indicated in the status bitmask
3924 * and the relevant info shall be in the error item.
3926 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3927 struct ib_mr_status *mr_status);
3929 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3930 u16 pkey, const union ib_gid *gid,
3931 const struct sockaddr *addr);
3932 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3933 struct ib_wq_init_attr *init_attr);
3934 int ib_destroy_wq(struct ib_wq *wq);
3935 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3937 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3938 struct ib_rwq_ind_table_init_attr*
3939 wq_ind_table_init_attr);
3940 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3942 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3943 unsigned int *sg_offset, unsigned int page_size);
3946 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3947 unsigned int *sg_offset, unsigned int page_size)
3951 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3957 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3958 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3960 void ib_drain_rq(struct ib_qp *qp);
3961 void ib_drain_sq(struct ib_qp *qp);
3962 void ib_drain_qp(struct ib_qp *qp);
3964 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3966 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3968 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3969 return attr->roce.dmac;
3973 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3975 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3976 attr->ib.dlid = (u16)dlid;
3977 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3978 attr->opa.dlid = dlid;
3981 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3983 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3984 return attr->ib.dlid;
3985 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3986 return attr->opa.dlid;
3990 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3995 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4000 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4003 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4004 attr->ib.src_path_bits = src_path_bits;
4005 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4006 attr->opa.src_path_bits = src_path_bits;
4009 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4011 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4012 return attr->ib.src_path_bits;
4013 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4014 return attr->opa.src_path_bits;
4018 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4021 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4022 attr->opa.make_grd = make_grd;
4025 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4027 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4028 return attr->opa.make_grd;
4032 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4034 attr->port_num = port_num;
4037 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4039 return attr->port_num;
4042 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4045 attr->static_rate = static_rate;
4048 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4050 return attr->static_rate;
4053 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4054 enum ib_ah_flags flag)
4056 attr->ah_flags = flag;
4059 static inline enum ib_ah_flags
4060 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4062 return attr->ah_flags;
4065 static inline const struct ib_global_route
4066 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4071 /*To retrieve and modify the grh */
4072 static inline struct ib_global_route
4073 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4078 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4080 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4082 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4085 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4088 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4090 grh->dgid.global.subnet_prefix = prefix;
4093 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4096 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4098 grh->dgid.global.interface_id = if_id;
4101 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4102 union ib_gid *dgid, u32 flow_label,
4103 u8 sgid_index, u8 hop_limit,
4106 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4108 attr->ah_flags = IB_AH_GRH;
4111 grh->flow_label = flow_label;
4112 grh->sgid_index = sgid_index;
4113 grh->hop_limit = hop_limit;
4114 grh->traffic_class = traffic_class;
4115 grh->sgid_attr = NULL;
4118 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4119 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4120 u32 flow_label, u8 hop_limit, u8 traffic_class,
4121 const struct ib_gid_attr *sgid_attr);
4122 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4123 const struct rdma_ah_attr *src);
4124 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4125 const struct rdma_ah_attr *new);
4126 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4129 * rdma_ah_find_type - Return address handle type.
4131 * @dev: Device to be checked
4132 * @port_num: Port number
4134 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4137 if (rdma_protocol_roce(dev, port_num))
4138 return RDMA_AH_ATTR_TYPE_ROCE;
4139 if (rdma_protocol_ib(dev, port_num)) {
4140 if (rdma_cap_opa_ah(dev, port_num))
4141 return RDMA_AH_ATTR_TYPE_OPA;
4142 return RDMA_AH_ATTR_TYPE_IB;
4145 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4149 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4150 * In the current implementation the only way to get
4151 * get the 32bit lid is from other sources for OPA.
4152 * For IB, lids will always be 16bits so cast the
4153 * value accordingly.
4157 static inline u16 ib_lid_cpu16(u32 lid)
4159 WARN_ON_ONCE(lid & 0xFFFF0000);
4164 * ib_lid_be16 - Return lid in 16bit BE encoding.
4168 static inline __be16 ib_lid_be16(u32 lid)
4170 WARN_ON_ONCE(lid & 0xFFFF0000);
4171 return cpu_to_be16((u16)lid);
4175 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4177 * @device: the rdma device
4178 * @comp_vector: index of completion vector
4180 * Returns NULL on failure, otherwise a corresponding cpu map of the
4181 * completion vector (returns all-cpus map if the device driver doesn't
4182 * implement get_vector_affinity).
4184 static inline const struct cpumask *
4185 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4187 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4188 !device->ops.get_vector_affinity)
4191 return device->ops.get_vector_affinity(device, comp_vector);
4196 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4197 * and add their gids, as needed, to the relevant RoCE devices.
4199 * @device: the rdma device
4201 void rdma_roce_rescan_device(struct ib_device *ibdev);
4203 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4206 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4208 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4209 enum rdma_netdev_t type, const char *name,
4210 unsigned char name_assign_type,
4211 void (*setup)(struct net_device *));
4213 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4214 enum rdma_netdev_t type, const char *name,
4215 unsigned char name_assign_type,
4216 void (*setup)(struct net_device *),
4217 struct net_device *netdev);
4220 * rdma_set_device_sysfs_group - Set device attributes group to have
4221 * driver specific sysfs entries at
4222 * for infiniband class.
4224 * @device: device pointer for which attributes to be created
4225 * @group: Pointer to group which should be added when device
4226 * is registered with sysfs.
4227 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4228 * group per device to have sysfs attributes.
4230 * NOTE: New drivers should not make use of this API; instead new device
4231 * parameter should be exposed via netlink command. This API and mechanism
4232 * exist only for existing drivers.
4235 rdma_set_device_sysfs_group(struct ib_device *dev,
4236 const struct attribute_group *group)
4238 dev->groups[1] = group;
4241 #endif /* IB_VERBS_H */