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),
241 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
244 enum ib_signature_prot_cap {
245 IB_PROT_T10DIF_TYPE_1 = 1,
246 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
247 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
250 enum ib_signature_guard_cap {
251 IB_GUARD_T10DIF_CRC = 1,
252 IB_GUARD_T10DIF_CSUM = 1 << 1,
261 enum ib_odp_general_cap_bits {
262 IB_ODP_SUPPORT = 1 << 0,
263 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
266 enum ib_odp_transport_cap_bits {
267 IB_ODP_SUPPORT_SEND = 1 << 0,
268 IB_ODP_SUPPORT_RECV = 1 << 1,
269 IB_ODP_SUPPORT_WRITE = 1 << 2,
270 IB_ODP_SUPPORT_READ = 1 << 3,
271 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
272 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
276 uint64_t general_caps;
278 uint32_t rc_odp_caps;
279 uint32_t uc_odp_caps;
280 uint32_t ud_odp_caps;
281 uint32_t xrc_odp_caps;
282 } per_transport_caps;
286 /* Corresponding bit will be set if qp type from
287 * 'enum ib_qp_type' is supported, e.g.
288 * supported_qpts |= 1 << IB_QPT_UD
291 u32 max_rwq_indirection_tables;
292 u32 max_rwq_indirection_table_size;
295 enum ib_tm_cap_flags {
296 /* Support tag matching on RC transport */
297 IB_TM_CAP_RC = 1 << 0,
301 /* Max size of RNDV header */
302 u32 max_rndv_hdr_size;
303 /* Max number of entries in tag matching list */
305 /* From enum ib_tm_cap_flags */
307 /* Max number of outstanding list operations */
309 /* Max number of SGE in tag matching entry */
313 struct ib_cq_init_attr {
319 enum ib_cq_attr_mask {
320 IB_CQ_MODERATE = 1 << 0,
324 u16 max_cq_moderation_count;
325 u16 max_cq_moderation_period;
328 struct ib_dm_mr_attr {
334 struct ib_dm_alloc_attr {
340 struct ib_device_attr {
342 __be64 sys_image_guid;
350 u64 device_cap_flags;
361 int max_qp_init_rd_atom;
362 int max_ee_init_rd_atom;
363 enum ib_atomic_cap atomic_cap;
364 enum ib_atomic_cap masked_atomic_cap;
371 int max_mcast_qp_attach;
372 int max_total_mcast_qp_attach;
379 unsigned int max_fast_reg_page_list_len;
381 u8 local_ca_ack_delay;
384 struct ib_odp_caps odp_caps;
385 uint64_t timestamp_mask;
386 uint64_t hca_core_clock; /* in KHZ */
387 struct ib_rss_caps rss_caps;
389 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
390 struct ib_tm_caps tm_caps;
391 struct ib_cq_caps cq_caps;
403 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
406 case IB_MTU_256: return 256;
407 case IB_MTU_512: return 512;
408 case IB_MTU_1024: return 1024;
409 case IB_MTU_2048: return 2048;
410 case IB_MTU_4096: return 4096;
415 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
419 else if (mtu >= 2048)
421 else if (mtu >= 1024)
435 IB_PORT_ACTIVE_DEFER = 5
446 static inline int ib_width_enum_to_int(enum ib_port_width width)
449 case IB_WIDTH_1X: return 1;
450 case IB_WIDTH_2X: return 2;
451 case IB_WIDTH_4X: return 4;
452 case IB_WIDTH_8X: return 8;
453 case IB_WIDTH_12X: return 12;
469 * struct rdma_hw_stats
470 * @lock - Mutex to protect parallel write access to lifespan and values
471 * of counters, which are 64bits and not guaranteeed to be written
472 * atomicaly on 32bits systems.
473 * @timestamp - Used by the core code to track when the last update was
474 * @lifespan - Used by the core code to determine how old the counters
475 * should be before being updated again. Stored in jiffies, defaults
476 * to 10 milliseconds, drivers can override the default be specifying
477 * their own value during their allocation routine.
478 * @name - Array of pointers to static names used for the counters in
480 * @num_counters - How many hardware counters there are. If name is
481 * shorter than this number, a kernel oops will result. Driver authors
482 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
483 * in their code to prevent this.
484 * @value - Array of u64 counters that are accessed by the sysfs code and
485 * filled in by the drivers get_stats routine
487 struct rdma_hw_stats {
488 struct mutex lock; /* Protect lifespan and values[] */
489 unsigned long timestamp;
490 unsigned long lifespan;
491 const char * const *names;
496 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
498 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
500 * @names - Array of static const char *
501 * @num_counters - How many elements in array
502 * @lifespan - How many milliseconds between updates
504 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
505 const char * const *names, int num_counters,
506 unsigned long lifespan)
508 struct rdma_hw_stats *stats;
510 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
514 stats->names = names;
515 stats->num_counters = num_counters;
516 stats->lifespan = msecs_to_jiffies(lifespan);
522 /* Define bits for the various functionality this port needs to be supported by
525 /* Management 0x00000FFF */
526 #define RDMA_CORE_CAP_IB_MAD 0x00000001
527 #define RDMA_CORE_CAP_IB_SMI 0x00000002
528 #define RDMA_CORE_CAP_IB_CM 0x00000004
529 #define RDMA_CORE_CAP_IW_CM 0x00000008
530 #define RDMA_CORE_CAP_IB_SA 0x00000010
531 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
533 /* Address format 0x000FF000 */
534 #define RDMA_CORE_CAP_AF_IB 0x00001000
535 #define RDMA_CORE_CAP_ETH_AH 0x00002000
536 #define RDMA_CORE_CAP_OPA_AH 0x00004000
537 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
539 /* Protocol 0xFFF00000 */
540 #define RDMA_CORE_CAP_PROT_IB 0x00100000
541 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
542 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
543 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
544 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
545 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
547 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
548 | RDMA_CORE_CAP_PROT_ROCE \
549 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
551 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
552 | RDMA_CORE_CAP_IB_MAD \
553 | RDMA_CORE_CAP_IB_SMI \
554 | RDMA_CORE_CAP_IB_CM \
555 | RDMA_CORE_CAP_IB_SA \
556 | RDMA_CORE_CAP_AF_IB)
557 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
558 | RDMA_CORE_CAP_IB_MAD \
559 | RDMA_CORE_CAP_IB_CM \
560 | RDMA_CORE_CAP_AF_IB \
561 | RDMA_CORE_CAP_ETH_AH)
562 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
563 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
564 | RDMA_CORE_CAP_IB_MAD \
565 | RDMA_CORE_CAP_IB_CM \
566 | RDMA_CORE_CAP_AF_IB \
567 | RDMA_CORE_CAP_ETH_AH)
568 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
569 | RDMA_CORE_CAP_IW_CM)
570 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
571 | RDMA_CORE_CAP_OPA_MAD)
573 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
575 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
577 struct ib_port_attr {
579 enum ib_port_state state;
581 enum ib_mtu active_mtu;
583 unsigned int ip_gids:1;
584 /* This is the value from PortInfo CapabilityMask, defined by IBA */
603 enum ib_device_modify_flags {
604 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
605 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
608 #define IB_DEVICE_NODE_DESC_MAX 64
610 struct ib_device_modify {
612 char node_desc[IB_DEVICE_NODE_DESC_MAX];
615 enum ib_port_modify_flags {
616 IB_PORT_SHUTDOWN = 1,
617 IB_PORT_INIT_TYPE = (1<<2),
618 IB_PORT_RESET_QKEY_CNTR = (1<<3),
619 IB_PORT_OPA_MASK_CHG = (1<<4)
622 struct ib_port_modify {
623 u32 set_port_cap_mask;
624 u32 clr_port_cap_mask;
632 IB_EVENT_QP_ACCESS_ERR,
636 IB_EVENT_PATH_MIG_ERR,
637 IB_EVENT_DEVICE_FATAL,
638 IB_EVENT_PORT_ACTIVE,
641 IB_EVENT_PKEY_CHANGE,
644 IB_EVENT_SRQ_LIMIT_REACHED,
645 IB_EVENT_QP_LAST_WQE_REACHED,
646 IB_EVENT_CLIENT_REREGISTER,
651 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
654 struct ib_device *device;
662 enum ib_event_type event;
665 struct ib_event_handler {
666 struct ib_device *device;
667 void (*handler)(struct ib_event_handler *, struct ib_event *);
668 struct list_head list;
671 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
673 (_ptr)->device = _device; \
674 (_ptr)->handler = _handler; \
675 INIT_LIST_HEAD(&(_ptr)->list); \
678 struct ib_global_route {
679 const struct ib_gid_attr *sgid_attr;
688 __be32 version_tclass_flow;
696 union rdma_network_hdr {
699 /* The IB spec states that if it's IPv4, the header
700 * is located in the last 20 bytes of the header.
703 struct iphdr roce4grh;
707 #define IB_QPN_MASK 0xFFFFFF
710 IB_MULTICAST_QPN = 0xffffff
713 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
714 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
721 IB_RATE_PORT_CURRENT = 0,
722 IB_RATE_2_5_GBPS = 2,
730 IB_RATE_120_GBPS = 10,
731 IB_RATE_14_GBPS = 11,
732 IB_RATE_56_GBPS = 12,
733 IB_RATE_112_GBPS = 13,
734 IB_RATE_168_GBPS = 14,
735 IB_RATE_25_GBPS = 15,
736 IB_RATE_100_GBPS = 16,
737 IB_RATE_200_GBPS = 17,
738 IB_RATE_300_GBPS = 18,
739 IB_RATE_28_GBPS = 19,
740 IB_RATE_50_GBPS = 20,
741 IB_RATE_400_GBPS = 21,
742 IB_RATE_600_GBPS = 22,
746 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
747 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
748 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
749 * @rate: rate to convert.
751 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
754 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
755 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
756 * @rate: rate to convert.
758 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
762 * enum ib_mr_type - memory region type
763 * @IB_MR_TYPE_MEM_REG: memory region that is used for
764 * normal registration
765 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
766 * signature operations (data-integrity
768 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
769 * register any arbitrary sg lists (without
770 * the normal mr constraints - see
775 IB_MR_TYPE_SIGNATURE,
781 * IB_SIG_TYPE_NONE: Unprotected.
782 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
784 enum ib_signature_type {
790 * Signature T10-DIF block-guard types
791 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
792 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
794 enum ib_t10_dif_bg_type {
800 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
802 * @bg_type: T10-DIF block guard type (CRC|CSUM)
803 * @pi_interval: protection information interval.
804 * @bg: seed of guard computation.
805 * @app_tag: application tag of guard block
806 * @ref_tag: initial guard block reference tag.
807 * @ref_remap: Indicate wethear the reftag increments each block
808 * @app_escape: Indicate to skip block check if apptag=0xffff
809 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
810 * @apptag_check_mask: check bitmask of application tag.
812 struct ib_t10_dif_domain {
813 enum ib_t10_dif_bg_type bg_type;
821 u16 apptag_check_mask;
825 * struct ib_sig_domain - Parameters for signature domain
826 * @sig_type: specific signauture type
827 * @sig: union of all signature domain attributes that may
828 * be used to set domain layout.
830 struct ib_sig_domain {
831 enum ib_signature_type sig_type;
833 struct ib_t10_dif_domain dif;
838 * struct ib_sig_attrs - Parameters for signature handover operation
839 * @check_mask: bitmask for signature byte check (8 bytes)
840 * @mem: memory domain layout desciptor.
841 * @wire: wire domain layout desciptor.
843 struct ib_sig_attrs {
845 struct ib_sig_domain mem;
846 struct ib_sig_domain wire;
849 enum ib_sig_err_type {
856 * Signature check masks (8 bytes in total) according to the T10-PI standard:
857 * -------- -------- ------------
858 * | GUARD | APPTAG | REFTAG |
860 * -------- -------- ------------
863 IB_SIG_CHECK_GUARD = 0xc0,
864 IB_SIG_CHECK_APPTAG = 0x30,
865 IB_SIG_CHECK_REFTAG = 0x0f,
869 * struct ib_sig_err - signature error descriptor
872 enum ib_sig_err_type err_type;
879 enum ib_mr_status_check {
880 IB_MR_CHECK_SIG_STATUS = 1,
884 * struct ib_mr_status - Memory region status container
886 * @fail_status: Bitmask of MR checks status. For each
887 * failed check a corresponding status bit is set.
888 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
891 struct ib_mr_status {
893 struct ib_sig_err sig_err;
897 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
899 * @mult: multiple to convert.
901 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
903 enum rdma_ah_attr_type {
904 RDMA_AH_ATTR_TYPE_UNDEFINED,
905 RDMA_AH_ATTR_TYPE_IB,
906 RDMA_AH_ATTR_TYPE_ROCE,
907 RDMA_AH_ATTR_TYPE_OPA,
915 struct roce_ah_attr {
925 struct rdma_ah_attr {
926 struct ib_global_route grh;
931 enum rdma_ah_attr_type type;
933 struct ib_ah_attr ib;
934 struct roce_ah_attr roce;
935 struct opa_ah_attr opa;
943 IB_WC_LOC_EEC_OP_ERR,
948 IB_WC_LOC_ACCESS_ERR,
949 IB_WC_REM_INV_REQ_ERR,
950 IB_WC_REM_ACCESS_ERR,
953 IB_WC_RNR_RETRY_EXC_ERR,
954 IB_WC_LOC_RDD_VIOL_ERR,
955 IB_WC_REM_INV_RD_REQ_ERR,
958 IB_WC_INV_EEC_STATE_ERR,
960 IB_WC_RESP_TIMEOUT_ERR,
964 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
975 IB_WC_MASKED_COMP_SWAP,
976 IB_WC_MASKED_FETCH_ADD,
978 * Set value of IB_WC_RECV so consumers can test if a completion is a
979 * receive by testing (opcode & IB_WC_RECV).
982 IB_WC_RECV_RDMA_WITH_IMM
987 IB_WC_WITH_IMM = (1<<1),
988 IB_WC_WITH_INVALIDATE = (1<<2),
989 IB_WC_IP_CSUM_OK = (1<<3),
990 IB_WC_WITH_SMAC = (1<<4),
991 IB_WC_WITH_VLAN = (1<<5),
992 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
998 struct ib_cqe *wr_cqe;
1000 enum ib_wc_status status;
1001 enum ib_wc_opcode opcode;
1007 u32 invalidate_rkey;
1015 u8 port_num; /* valid only for DR SMPs on switches */
1018 u8 network_hdr_type;
1021 enum ib_cq_notify_flags {
1022 IB_CQ_SOLICITED = 1 << 0,
1023 IB_CQ_NEXT_COMP = 1 << 1,
1024 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1025 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1034 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1036 return srq_type == IB_SRQT_XRC ||
1037 srq_type == IB_SRQT_TM;
1040 enum ib_srq_attr_mask {
1041 IB_SRQ_MAX_WR = 1 << 0,
1042 IB_SRQ_LIMIT = 1 << 1,
1045 struct ib_srq_attr {
1051 struct ib_srq_init_attr {
1052 void (*event_handler)(struct ib_event *, void *);
1054 struct ib_srq_attr attr;
1055 enum ib_srq_type srq_type;
1061 struct ib_xrcd *xrcd;
1076 u32 max_inline_data;
1079 * Maximum number of rdma_rw_ctx structures in flight at a time.
1080 * ib_create_qp() will calculate the right amount of neededed WRs
1081 * and MRs based on this.
1093 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1094 * here (and in that order) since the MAD layer uses them as
1095 * indices into a 2-entry table.
1104 IB_QPT_RAW_ETHERTYPE,
1105 IB_QPT_RAW_PACKET = 8,
1109 IB_QPT_DRIVER = 0xFF,
1110 /* Reserve a range for qp types internal to the low level driver.
1111 * These qp types will not be visible at the IB core layer, so the
1112 * IB_QPT_MAX usages should not be affected in the core layer
1114 IB_QPT_RESERVED1 = 0x1000,
1126 enum ib_qp_create_flags {
1127 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1128 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1129 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1130 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1131 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1132 IB_QP_CREATE_NETIF_QP = 1 << 5,
1133 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1134 /* FREE = 1 << 7, */
1135 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1136 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1137 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1138 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1139 /* reserve bits 26-31 for low level drivers' internal use */
1140 IB_QP_CREATE_RESERVED_START = 1 << 26,
1141 IB_QP_CREATE_RESERVED_END = 1 << 31,
1145 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1146 * callback to destroy the passed in QP.
1149 struct ib_qp_init_attr {
1150 /* Consumer's event_handler callback must not block */
1151 void (*event_handler)(struct ib_event *, void *);
1154 struct ib_cq *send_cq;
1155 struct ib_cq *recv_cq;
1157 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1158 struct ib_qp_cap cap;
1159 enum ib_sig_type sq_sig_type;
1160 enum ib_qp_type qp_type;
1164 * Only needed for special QP types, or when using the RW API.
1167 struct ib_rwq_ind_table *rwq_ind_tbl;
1171 struct ib_qp_open_attr {
1172 void (*event_handler)(struct ib_event *, void *);
1175 enum ib_qp_type qp_type;
1178 enum ib_rnr_timeout {
1179 IB_RNR_TIMER_655_36 = 0,
1180 IB_RNR_TIMER_000_01 = 1,
1181 IB_RNR_TIMER_000_02 = 2,
1182 IB_RNR_TIMER_000_03 = 3,
1183 IB_RNR_TIMER_000_04 = 4,
1184 IB_RNR_TIMER_000_06 = 5,
1185 IB_RNR_TIMER_000_08 = 6,
1186 IB_RNR_TIMER_000_12 = 7,
1187 IB_RNR_TIMER_000_16 = 8,
1188 IB_RNR_TIMER_000_24 = 9,
1189 IB_RNR_TIMER_000_32 = 10,
1190 IB_RNR_TIMER_000_48 = 11,
1191 IB_RNR_TIMER_000_64 = 12,
1192 IB_RNR_TIMER_000_96 = 13,
1193 IB_RNR_TIMER_001_28 = 14,
1194 IB_RNR_TIMER_001_92 = 15,
1195 IB_RNR_TIMER_002_56 = 16,
1196 IB_RNR_TIMER_003_84 = 17,
1197 IB_RNR_TIMER_005_12 = 18,
1198 IB_RNR_TIMER_007_68 = 19,
1199 IB_RNR_TIMER_010_24 = 20,
1200 IB_RNR_TIMER_015_36 = 21,
1201 IB_RNR_TIMER_020_48 = 22,
1202 IB_RNR_TIMER_030_72 = 23,
1203 IB_RNR_TIMER_040_96 = 24,
1204 IB_RNR_TIMER_061_44 = 25,
1205 IB_RNR_TIMER_081_92 = 26,
1206 IB_RNR_TIMER_122_88 = 27,
1207 IB_RNR_TIMER_163_84 = 28,
1208 IB_RNR_TIMER_245_76 = 29,
1209 IB_RNR_TIMER_327_68 = 30,
1210 IB_RNR_TIMER_491_52 = 31
1213 enum ib_qp_attr_mask {
1215 IB_QP_CUR_STATE = (1<<1),
1216 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1217 IB_QP_ACCESS_FLAGS = (1<<3),
1218 IB_QP_PKEY_INDEX = (1<<4),
1219 IB_QP_PORT = (1<<5),
1220 IB_QP_QKEY = (1<<6),
1222 IB_QP_PATH_MTU = (1<<8),
1223 IB_QP_TIMEOUT = (1<<9),
1224 IB_QP_RETRY_CNT = (1<<10),
1225 IB_QP_RNR_RETRY = (1<<11),
1226 IB_QP_RQ_PSN = (1<<12),
1227 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1228 IB_QP_ALT_PATH = (1<<14),
1229 IB_QP_MIN_RNR_TIMER = (1<<15),
1230 IB_QP_SQ_PSN = (1<<16),
1231 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1232 IB_QP_PATH_MIG_STATE = (1<<18),
1233 IB_QP_CAP = (1<<19),
1234 IB_QP_DEST_QPN = (1<<20),
1235 IB_QP_RESERVED1 = (1<<21),
1236 IB_QP_RESERVED2 = (1<<22),
1237 IB_QP_RESERVED3 = (1<<23),
1238 IB_QP_RESERVED4 = (1<<24),
1239 IB_QP_RATE_LIMIT = (1<<25),
1264 enum ib_qp_state qp_state;
1265 enum ib_qp_state cur_qp_state;
1266 enum ib_mtu path_mtu;
1267 enum ib_mig_state path_mig_state;
1272 int qp_access_flags;
1273 struct ib_qp_cap cap;
1274 struct rdma_ah_attr ah_attr;
1275 struct rdma_ah_attr alt_ah_attr;
1278 u8 en_sqd_async_notify;
1281 u8 max_dest_rd_atomic;
1293 /* These are shared with userspace */
1294 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1295 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1296 IB_WR_SEND = IB_UVERBS_WR_SEND,
1297 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1298 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1299 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1300 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1301 IB_WR_LSO = IB_UVERBS_WR_TSO,
1302 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1303 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1304 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1305 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1306 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1307 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1308 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1310 /* These are kernel only and can not be issued by userspace */
1311 IB_WR_REG_MR = 0x20,
1314 /* reserve values for low level drivers' internal use.
1315 * These values will not be used at all in the ib core layer.
1317 IB_WR_RESERVED1 = 0xf0,
1329 enum ib_send_flags {
1331 IB_SEND_SIGNALED = (1<<1),
1332 IB_SEND_SOLICITED = (1<<2),
1333 IB_SEND_INLINE = (1<<3),
1334 IB_SEND_IP_CSUM = (1<<4),
1336 /* reserve bits 26-31 for low level drivers' internal use */
1337 IB_SEND_RESERVED_START = (1 << 26),
1338 IB_SEND_RESERVED_END = (1 << 31),
1348 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1352 struct ib_send_wr *next;
1355 struct ib_cqe *wr_cqe;
1357 struct ib_sge *sg_list;
1359 enum ib_wr_opcode opcode;
1363 u32 invalidate_rkey;
1368 struct ib_send_wr wr;
1373 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1375 return container_of(wr, struct ib_rdma_wr, wr);
1378 struct ib_atomic_wr {
1379 struct ib_send_wr wr;
1383 u64 compare_add_mask;
1388 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1390 return container_of(wr, struct ib_atomic_wr, wr);
1394 struct ib_send_wr wr;
1401 u16 pkey_index; /* valid for GSI only */
1402 u8 port_num; /* valid for DR SMPs on switch only */
1405 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1407 return container_of(wr, struct ib_ud_wr, wr);
1411 struct ib_send_wr wr;
1417 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1419 return container_of(wr, struct ib_reg_wr, wr);
1422 struct ib_sig_handover_wr {
1423 struct ib_send_wr wr;
1424 struct ib_sig_attrs *sig_attrs;
1425 struct ib_mr *sig_mr;
1427 struct ib_sge *prot;
1430 static inline const struct ib_sig_handover_wr *
1431 sig_handover_wr(const struct ib_send_wr *wr)
1433 return container_of(wr, struct ib_sig_handover_wr, wr);
1437 struct ib_recv_wr *next;
1440 struct ib_cqe *wr_cqe;
1442 struct ib_sge *sg_list;
1446 enum ib_access_flags {
1447 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1448 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1449 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1450 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1451 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1452 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1453 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1454 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1456 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1460 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1461 * are hidden here instead of a uapi header!
1463 enum ib_mr_rereg_flags {
1464 IB_MR_REREG_TRANS = 1,
1465 IB_MR_REREG_PD = (1<<1),
1466 IB_MR_REREG_ACCESS = (1<<2),
1467 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1470 struct ib_fmr_attr {
1478 enum rdma_remove_reason {
1480 * Userspace requested uobject deletion or initial try
1481 * to remove uobject via cleanup. Call could fail
1483 RDMA_REMOVE_DESTROY,
1484 /* Context deletion. This call should delete the actual object itself */
1486 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1487 RDMA_REMOVE_DRIVER_REMOVE,
1488 /* uobj is being cleaned-up before being committed */
1492 struct ib_rdmacg_object {
1493 #ifdef CONFIG_CGROUP_RDMA
1494 struct rdma_cgroup *cg; /* owner rdma cgroup */
1498 struct ib_ucontext {
1499 struct ib_device *device;
1500 struct ib_uverbs_file *ufile;
1502 * 'closing' can be read by the driver only during a destroy callback,
1503 * it is set when we are closing the file descriptor and indicates
1504 * that mm_sem may be locked.
1508 bool cleanup_retryable;
1510 void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1511 unsigned long start, unsigned long end);
1512 struct mutex per_mm_list_lock;
1513 struct list_head per_mm_list;
1515 struct ib_rdmacg_object cg_obj;
1517 * Implementation details of the RDMA core, don't use in drivers:
1519 struct rdma_restrack_entry res;
1523 u64 user_handle; /* handle given to us by userspace */
1524 /* ufile & ucontext owning this object */
1525 struct ib_uverbs_file *ufile;
1526 /* FIXME, save memory: ufile->context == context */
1527 struct ib_ucontext *context; /* associated user context */
1528 void *object; /* containing object */
1529 struct list_head list; /* link to context's list */
1530 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1531 int id; /* index into kernel idr */
1533 atomic_t usecnt; /* protects exclusive access */
1534 struct rcu_head rcu; /* kfree_rcu() overhead */
1536 const struct uverbs_api_object *uapi_object;
1540 const void __user *inbuf;
1541 void __user *outbuf;
1549 struct ib_device *device;
1550 struct ib_uobject *uobject;
1551 atomic_t usecnt; /* count all resources */
1553 u32 unsafe_global_rkey;
1556 * Implementation details of the RDMA core, don't use in drivers:
1558 struct ib_mr *__internal_mr;
1559 struct rdma_restrack_entry res;
1563 struct ib_device *device;
1564 atomic_t usecnt; /* count all exposed resources */
1565 struct inode *inode;
1567 struct mutex tgt_qp_mutex;
1568 struct list_head tgt_qp_list;
1572 struct ib_device *device;
1574 struct ib_uobject *uobject;
1575 const struct ib_gid_attr *sgid_attr;
1576 enum rdma_ah_attr_type type;
1579 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1581 enum ib_poll_context {
1582 IB_POLL_DIRECT, /* caller context, no hw completions */
1583 IB_POLL_SOFTIRQ, /* poll from softirq context */
1584 IB_POLL_WORKQUEUE, /* poll from workqueue */
1585 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1589 struct ib_device *device;
1590 struct ib_uobject *uobject;
1591 ib_comp_handler comp_handler;
1592 void (*event_handler)(struct ib_event *, void *);
1595 atomic_t usecnt; /* count number of work queues */
1596 enum ib_poll_context poll_ctx;
1599 struct irq_poll iop;
1600 struct work_struct work;
1602 struct workqueue_struct *comp_wq;
1604 * Implementation details of the RDMA core, don't use in drivers:
1606 struct rdma_restrack_entry res;
1610 struct ib_device *device;
1612 struct ib_uobject *uobject;
1613 void (*event_handler)(struct ib_event *, void *);
1615 enum ib_srq_type srq_type;
1622 struct ib_xrcd *xrcd;
1629 enum ib_raw_packet_caps {
1630 /* Strip cvlan from incoming packet and report it in the matching work
1631 * completion is supported.
1633 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1634 /* Scatter FCS field of an incoming packet to host memory is supported.
1636 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1637 /* Checksum offloads are supported (for both send and receive). */
1638 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1639 /* When a packet is received for an RQ with no receive WQEs, the
1640 * packet processing is delayed.
1642 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1656 struct ib_device *device;
1657 struct ib_uobject *uobject;
1659 void (*event_handler)(struct ib_event *, void *);
1663 enum ib_wq_state state;
1664 enum ib_wq_type wq_type;
1669 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1670 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1671 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1672 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1675 struct ib_wq_init_attr {
1677 enum ib_wq_type wq_type;
1681 void (*event_handler)(struct ib_event *, void *);
1682 u32 create_flags; /* Use enum ib_wq_flags */
1685 enum ib_wq_attr_mask {
1686 IB_WQ_STATE = 1 << 0,
1687 IB_WQ_CUR_STATE = 1 << 1,
1688 IB_WQ_FLAGS = 1 << 2,
1692 enum ib_wq_state wq_state;
1693 enum ib_wq_state curr_wq_state;
1694 u32 flags; /* Use enum ib_wq_flags */
1695 u32 flags_mask; /* Use enum ib_wq_flags */
1698 struct ib_rwq_ind_table {
1699 struct ib_device *device;
1700 struct ib_uobject *uobject;
1703 u32 log_ind_tbl_size;
1704 struct ib_wq **ind_tbl;
1707 struct ib_rwq_ind_table_init_attr {
1708 u32 log_ind_tbl_size;
1709 /* Each entry is a pointer to Receive Work Queue */
1710 struct ib_wq **ind_tbl;
1713 enum port_pkey_state {
1714 IB_PORT_PKEY_NOT_VALID = 0,
1715 IB_PORT_PKEY_VALID = 1,
1716 IB_PORT_PKEY_LISTED = 2,
1719 struct ib_qp_security;
1721 struct ib_port_pkey {
1722 enum port_pkey_state state;
1725 struct list_head qp_list;
1726 struct list_head to_error_list;
1727 struct ib_qp_security *sec;
1730 struct ib_ports_pkeys {
1731 struct ib_port_pkey main;
1732 struct ib_port_pkey alt;
1735 struct ib_qp_security {
1737 struct ib_device *dev;
1738 /* Hold this mutex when changing port and pkey settings. */
1740 struct ib_ports_pkeys *ports_pkeys;
1741 /* A list of all open shared QP handles. Required to enforce security
1742 * properly for all users of a shared QP.
1744 struct list_head shared_qp_list;
1747 atomic_t error_list_count;
1748 struct completion error_complete;
1749 int error_comps_pending;
1753 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1754 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1757 struct ib_device *device;
1759 struct ib_cq *send_cq;
1760 struct ib_cq *recv_cq;
1763 struct list_head rdma_mrs;
1764 struct list_head sig_mrs;
1766 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1767 struct list_head xrcd_list;
1769 /* count times opened, mcast attaches, flow attaches */
1771 struct list_head open_list;
1772 struct ib_qp *real_qp;
1773 struct ib_uobject *uobject;
1774 void (*event_handler)(struct ib_event *, void *);
1776 /* sgid_attrs associated with the AV's */
1777 const struct ib_gid_attr *av_sgid_attr;
1778 const struct ib_gid_attr *alt_path_sgid_attr;
1782 enum ib_qp_type qp_type;
1783 struct ib_rwq_ind_table *rwq_ind_tbl;
1784 struct ib_qp_security *qp_sec;
1788 * Implementation details of the RDMA core, don't use in drivers:
1790 struct rdma_restrack_entry res;
1794 struct ib_device *device;
1797 struct ib_uobject *uobject;
1802 struct ib_device *device;
1808 unsigned int page_size;
1811 struct ib_uobject *uobject; /* user */
1812 struct list_head qp_entry; /* FR */
1818 * Implementation details of the RDMA core, don't use in drivers:
1820 struct rdma_restrack_entry res;
1824 struct ib_device *device;
1826 struct ib_uobject *uobject;
1828 enum ib_mw_type type;
1832 struct ib_device *device;
1834 struct list_head list;
1839 /* Supported steering options */
1840 enum ib_flow_attr_type {
1841 /* steering according to rule specifications */
1842 IB_FLOW_ATTR_NORMAL = 0x0,
1843 /* default unicast and multicast rule -
1844 * receive all Eth traffic which isn't steered to any QP
1846 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1847 /* default multicast rule -
1848 * receive all Eth multicast traffic which isn't steered to any QP
1850 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1851 /* sniffer rule - receive all port traffic */
1852 IB_FLOW_ATTR_SNIFFER = 0x3
1855 /* Supported steering header types */
1856 enum ib_flow_spec_type {
1858 IB_FLOW_SPEC_ETH = 0x20,
1859 IB_FLOW_SPEC_IB = 0x22,
1861 IB_FLOW_SPEC_IPV4 = 0x30,
1862 IB_FLOW_SPEC_IPV6 = 0x31,
1863 IB_FLOW_SPEC_ESP = 0x34,
1865 IB_FLOW_SPEC_TCP = 0x40,
1866 IB_FLOW_SPEC_UDP = 0x41,
1867 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1868 IB_FLOW_SPEC_GRE = 0x51,
1869 IB_FLOW_SPEC_MPLS = 0x60,
1870 IB_FLOW_SPEC_INNER = 0x100,
1872 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1873 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1874 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1875 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1877 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1878 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1880 /* Flow steering rule priority is set according to it's domain.
1881 * Lower domain value means higher priority.
1883 enum ib_flow_domain {
1884 IB_FLOW_DOMAIN_USER,
1885 IB_FLOW_DOMAIN_ETHTOOL,
1888 IB_FLOW_DOMAIN_NUM /* Must be last */
1891 enum ib_flow_flags {
1892 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1893 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1894 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1897 struct ib_flow_eth_filter {
1906 struct ib_flow_spec_eth {
1909 struct ib_flow_eth_filter val;
1910 struct ib_flow_eth_filter mask;
1913 struct ib_flow_ib_filter {
1920 struct ib_flow_spec_ib {
1923 struct ib_flow_ib_filter val;
1924 struct ib_flow_ib_filter mask;
1927 /* IPv4 header flags */
1928 enum ib_ipv4_flags {
1929 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1930 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1931 last have this flag set */
1934 struct ib_flow_ipv4_filter {
1945 struct ib_flow_spec_ipv4 {
1948 struct ib_flow_ipv4_filter val;
1949 struct ib_flow_ipv4_filter mask;
1952 struct ib_flow_ipv6_filter {
1963 struct ib_flow_spec_ipv6 {
1966 struct ib_flow_ipv6_filter val;
1967 struct ib_flow_ipv6_filter mask;
1970 struct ib_flow_tcp_udp_filter {
1977 struct ib_flow_spec_tcp_udp {
1980 struct ib_flow_tcp_udp_filter val;
1981 struct ib_flow_tcp_udp_filter mask;
1984 struct ib_flow_tunnel_filter {
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990 * the tunnel_id from val has the vni value
1992 struct ib_flow_spec_tunnel {
1995 struct ib_flow_tunnel_filter val;
1996 struct ib_flow_tunnel_filter mask;
1999 struct ib_flow_esp_filter {
2006 struct ib_flow_spec_esp {
2009 struct ib_flow_esp_filter val;
2010 struct ib_flow_esp_filter mask;
2013 struct ib_flow_gre_filter {
2014 __be16 c_ks_res0_ver;
2021 struct ib_flow_spec_gre {
2024 struct ib_flow_gre_filter val;
2025 struct ib_flow_gre_filter mask;
2028 struct ib_flow_mpls_filter {
2034 struct ib_flow_spec_mpls {
2037 struct ib_flow_mpls_filter val;
2038 struct ib_flow_mpls_filter mask;
2041 struct ib_flow_spec_action_tag {
2042 enum ib_flow_spec_type type;
2047 struct ib_flow_spec_action_drop {
2048 enum ib_flow_spec_type type;
2052 struct ib_flow_spec_action_handle {
2053 enum ib_flow_spec_type type;
2055 struct ib_flow_action *act;
2058 enum ib_counters_description {
2063 struct ib_flow_spec_action_count {
2064 enum ib_flow_spec_type type;
2066 struct ib_counters *counters;
2069 union ib_flow_spec {
2074 struct ib_flow_spec_eth eth;
2075 struct ib_flow_spec_ib ib;
2076 struct ib_flow_spec_ipv4 ipv4;
2077 struct ib_flow_spec_tcp_udp tcp_udp;
2078 struct ib_flow_spec_ipv6 ipv6;
2079 struct ib_flow_spec_tunnel tunnel;
2080 struct ib_flow_spec_esp esp;
2081 struct ib_flow_spec_gre gre;
2082 struct ib_flow_spec_mpls mpls;
2083 struct ib_flow_spec_action_tag flow_tag;
2084 struct ib_flow_spec_action_drop drop;
2085 struct ib_flow_spec_action_handle action;
2086 struct ib_flow_spec_action_count flow_count;
2089 struct ib_flow_attr {
2090 enum ib_flow_attr_type type;
2096 union ib_flow_spec flows[];
2101 struct ib_device *device;
2102 struct ib_uobject *uobject;
2105 enum ib_flow_action_type {
2106 IB_FLOW_ACTION_UNSPECIFIED,
2107 IB_FLOW_ACTION_ESP = 1,
2110 struct ib_flow_action_attrs_esp_keymats {
2111 enum ib_uverbs_flow_action_esp_keymat protocol;
2113 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2117 struct ib_flow_action_attrs_esp_replays {
2118 enum ib_uverbs_flow_action_esp_replay protocol;
2120 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2124 enum ib_flow_action_attrs_esp_flags {
2125 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2126 * This is done in order to share the same flags between user-space and
2127 * kernel and spare an unnecessary translation.
2131 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2132 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2135 struct ib_flow_spec_list {
2136 struct ib_flow_spec_list *next;
2137 union ib_flow_spec spec;
2140 struct ib_flow_action_attrs_esp {
2141 struct ib_flow_action_attrs_esp_keymats *keymat;
2142 struct ib_flow_action_attrs_esp_replays *replay;
2143 struct ib_flow_spec_list *encap;
2144 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2145 * Value of 0 is a valid value.
2151 /* Use enum ib_flow_action_attrs_esp_flags */
2153 u64 hard_limit_pkts;
2156 struct ib_flow_action {
2157 struct ib_device *device;
2158 struct ib_uobject *uobject;
2159 enum ib_flow_action_type type;
2166 enum ib_process_mad_flags {
2167 IB_MAD_IGNORE_MKEY = 1,
2168 IB_MAD_IGNORE_BKEY = 2,
2169 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2172 enum ib_mad_result {
2173 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2174 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2175 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2176 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2179 struct ib_port_cache {
2181 struct ib_pkey_cache *pkey;
2182 struct ib_gid_table *gid;
2184 enum ib_port_state port_state;
2189 struct ib_event_handler event_handler;
2194 struct ib_port_immutable {
2201 struct ib_port_data {
2202 struct ib_device *ib_dev;
2204 struct ib_port_immutable immutable;
2206 spinlock_t pkey_list_lock;
2207 struct list_head pkey_list;
2209 struct ib_port_cache cache;
2211 spinlock_t netdev_lock;
2212 struct net_device __rcu *netdev;
2213 struct hlist_node ndev_hash_link;
2216 /* rdma netdev type - specifies protocol type */
2217 enum rdma_netdev_t {
2218 RDMA_NETDEV_OPA_VNIC,
2223 * struct rdma_netdev - rdma netdev
2224 * For cases where netstack interfacing is required.
2226 struct rdma_netdev {
2228 struct ib_device *hca;
2232 * cleanup function must be specified.
2233 * FIXME: This is only used for OPA_VNIC and that usage should be
2236 void (*free_rdma_netdev)(struct net_device *netdev);
2238 /* control functions */
2239 void (*set_id)(struct net_device *netdev, int id);
2241 int (*send)(struct net_device *dev, struct sk_buff *skb,
2242 struct ib_ah *address, u32 dqpn);
2244 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2245 union ib_gid *gid, u16 mlid,
2246 int set_qkey, u32 qkey);
2247 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2248 union ib_gid *gid, u16 mlid);
2251 struct rdma_netdev_alloc_params {
2257 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2258 struct net_device *netdev, void *param);
2261 struct ib_counters {
2262 struct ib_device *device;
2263 struct ib_uobject *uobject;
2264 /* num of objects attached */
2268 struct ib_counters_read_attr {
2271 u32 flags; /* use enum ib_read_counters_flags */
2274 struct uverbs_attr_bundle;
2276 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2277 .size_##ib_struct = \
2278 (sizeof(struct drv_struct) + \
2279 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2280 BUILD_BUG_ON_ZERO( \
2281 !__same_type(((struct drv_struct *)NULL)->member, \
2284 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2285 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, GFP_KERNEL))
2287 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2290 * struct ib_device_ops - InfiniBand device operations
2291 * This structure defines all the InfiniBand device operations, providers will
2292 * need to define the supported operations, otherwise they will be set to null.
2294 struct ib_device_ops {
2295 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2296 const struct ib_send_wr **bad_send_wr);
2297 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2298 const struct ib_recv_wr **bad_recv_wr);
2299 void (*drain_rq)(struct ib_qp *qp);
2300 void (*drain_sq)(struct ib_qp *qp);
2301 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2302 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2303 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2304 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2305 int (*post_srq_recv)(struct ib_srq *srq,
2306 const struct ib_recv_wr *recv_wr,
2307 const struct ib_recv_wr **bad_recv_wr);
2308 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2309 u8 port_num, const struct ib_wc *in_wc,
2310 const struct ib_grh *in_grh,
2311 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2312 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2313 u16 *out_mad_pkey_index);
2314 int (*query_device)(struct ib_device *device,
2315 struct ib_device_attr *device_attr,
2316 struct ib_udata *udata);
2317 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2318 struct ib_device_modify *device_modify);
2319 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2320 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2322 int (*query_port)(struct ib_device *device, u8 port_num,
2323 struct ib_port_attr *port_attr);
2324 int (*modify_port)(struct ib_device *device, u8 port_num,
2325 int port_modify_mask,
2326 struct ib_port_modify *port_modify);
2328 * The following mandatory functions are used only at device
2329 * registration. Keep functions such as these at the end of this
2330 * structure to avoid cache line misses when accessing struct ib_device
2333 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2334 struct ib_port_immutable *immutable);
2335 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2338 * When calling get_netdev, the HW vendor's driver should return the
2339 * net device of device @device at port @port_num or NULL if such
2340 * a net device doesn't exist. The vendor driver should call dev_hold
2341 * on this net device. The HW vendor's device driver must guarantee
2342 * that this function returns NULL before the net device has finished
2343 * NETDEV_UNREGISTER state.
2345 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2347 * rdma netdev operation
2349 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2350 * must return -EOPNOTSUPP if it doesn't support the specified type.
2352 struct net_device *(*alloc_rdma_netdev)(
2353 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2354 const char *name, unsigned char name_assign_type,
2355 void (*setup)(struct net_device *));
2357 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2358 enum rdma_netdev_t type,
2359 struct rdma_netdev_alloc_params *params);
2361 * query_gid should be return GID value for @device, when @port_num
2362 * link layer is either IB or iWarp. It is no-op if @port_num port
2363 * is RoCE link layer.
2365 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2368 * When calling add_gid, the HW vendor's driver should add the gid
2369 * of device of port at gid index available at @attr. Meta-info of
2370 * that gid (for example, the network device related to this gid) is
2371 * available at @attr. @context allows the HW vendor driver to store
2372 * extra information together with a GID entry. The HW vendor driver may
2373 * allocate memory to contain this information and store it in @context
2374 * when a new GID entry is written to. Params are consistent until the
2375 * next call of add_gid or delete_gid. The function should return 0 on
2376 * success or error otherwise. The function could be called
2377 * concurrently for different ports. This function is only called when
2378 * roce_gid_table is used.
2380 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2382 * When calling del_gid, the HW vendor's driver should delete the
2383 * gid of device @device at gid index gid_index of port port_num
2384 * available in @attr.
2385 * Upon the deletion of a GID entry, the HW vendor must free any
2386 * allocated memory. The caller will clear @context afterwards.
2387 * This function is only called when roce_gid_table is used.
2389 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2390 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2392 int (*alloc_ucontext)(struct ib_ucontext *context,
2393 struct ib_udata *udata);
2394 void (*dealloc_ucontext)(struct ib_ucontext *context);
2395 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2396 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2397 int (*alloc_pd)(struct ib_pd *pd, struct ib_ucontext *context,
2398 struct ib_udata *udata);
2399 void (*dealloc_pd)(struct ib_pd *pd);
2400 struct ib_ah *(*create_ah)(struct ib_pd *pd,
2401 struct rdma_ah_attr *ah_attr, u32 flags,
2402 struct ib_udata *udata);
2403 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2404 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2405 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2406 struct ib_srq *(*create_srq)(struct ib_pd *pd,
2407 struct ib_srq_init_attr *srq_init_attr,
2408 struct ib_udata *udata);
2409 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2410 enum ib_srq_attr_mask srq_attr_mask,
2411 struct ib_udata *udata);
2412 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2413 int (*destroy_srq)(struct ib_srq *srq);
2414 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2415 struct ib_qp_init_attr *qp_init_attr,
2416 struct ib_udata *udata);
2417 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2418 int qp_attr_mask, struct ib_udata *udata);
2419 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2420 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2421 int (*destroy_qp)(struct ib_qp *qp);
2422 struct ib_cq *(*create_cq)(struct ib_device *device,
2423 const struct ib_cq_init_attr *attr,
2424 struct ib_ucontext *context,
2425 struct ib_udata *udata);
2426 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2427 int (*destroy_cq)(struct ib_cq *cq);
2428 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2429 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2430 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2431 u64 virt_addr, int mr_access_flags,
2432 struct ib_udata *udata);
2433 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2434 u64 virt_addr, int mr_access_flags,
2435 struct ib_pd *pd, struct ib_udata *udata);
2436 int (*dereg_mr)(struct ib_mr *mr);
2437 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2439 int (*advise_mr)(struct ib_pd *pd,
2440 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2441 struct ib_sge *sg_list, u32 num_sge,
2442 struct uverbs_attr_bundle *attrs);
2443 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2444 unsigned int *sg_offset);
2445 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2446 struct ib_mr_status *mr_status);
2447 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2448 struct ib_udata *udata);
2449 int (*dealloc_mw)(struct ib_mw *mw);
2450 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2451 struct ib_fmr_attr *fmr_attr);
2452 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2454 int (*unmap_fmr)(struct list_head *fmr_list);
2455 int (*dealloc_fmr)(struct ib_fmr *fmr);
2456 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2457 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2458 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2459 struct ib_ucontext *ucontext,
2460 struct ib_udata *udata);
2461 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2462 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2463 struct ib_flow_attr *flow_attr,
2464 int domain, struct ib_udata *udata);
2465 int (*destroy_flow)(struct ib_flow *flow_id);
2466 struct ib_flow_action *(*create_flow_action_esp)(
2467 struct ib_device *device,
2468 const struct ib_flow_action_attrs_esp *attr,
2469 struct uverbs_attr_bundle *attrs);
2470 int (*destroy_flow_action)(struct ib_flow_action *action);
2471 int (*modify_flow_action_esp)(
2472 struct ib_flow_action *action,
2473 const struct ib_flow_action_attrs_esp *attr,
2474 struct uverbs_attr_bundle *attrs);
2475 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2477 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2478 struct ifla_vf_info *ivf);
2479 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2480 struct ifla_vf_stats *stats);
2481 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2483 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2484 struct ib_wq_init_attr *init_attr,
2485 struct ib_udata *udata);
2486 int (*destroy_wq)(struct ib_wq *wq);
2487 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2488 u32 wq_attr_mask, struct ib_udata *udata);
2489 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2490 struct ib_device *device,
2491 struct ib_rwq_ind_table_init_attr *init_attr,
2492 struct ib_udata *udata);
2493 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2494 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2495 struct ib_ucontext *context,
2496 struct ib_dm_alloc_attr *attr,
2497 struct uverbs_attr_bundle *attrs);
2498 int (*dealloc_dm)(struct ib_dm *dm);
2499 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2500 struct ib_dm_mr_attr *attr,
2501 struct uverbs_attr_bundle *attrs);
2502 struct ib_counters *(*create_counters)(
2503 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2504 int (*destroy_counters)(struct ib_counters *counters);
2505 int (*read_counters)(struct ib_counters *counters,
2506 struct ib_counters_read_attr *counters_read_attr,
2507 struct uverbs_attr_bundle *attrs);
2509 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2510 * driver initialized data. The struct is kfree()'ed by the sysfs
2511 * core when the device is removed. A lifespan of -1 in the return
2512 * struct tells the core to set a default lifespan.
2514 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2517 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2518 * @index - The index in the value array we wish to have updated, or
2519 * num_counters if we want all stats updated
2521 * < 0 - Error, no counters updated
2522 * index - Updated the single counter pointed to by index
2523 * num_counters - Updated all counters (will reset the timestamp
2524 * and prevent further calls for lifespan milliseconds)
2525 * Drivers are allowed to update all counters in leiu of just the
2526 * one given in index at their option
2528 int (*get_hw_stats)(struct ib_device *device,
2529 struct rdma_hw_stats *stats, u8 port, int index);
2531 * This function is called once for each port when a ib device is
2534 int (*init_port)(struct ib_device *device, u8 port_num,
2535 struct kobject *port_sysfs);
2537 * Allows rdma drivers to add their own restrack attributes.
2539 int (*fill_res_entry)(struct sk_buff *msg,
2540 struct rdma_restrack_entry *entry);
2542 /* Device lifecycle callbacks */
2544 * Called after the device becomes registered, before clients are
2547 int (*enable_driver)(struct ib_device *dev);
2549 * This is called as part of ib_dealloc_device().
2551 void (*dealloc_driver)(struct ib_device *dev);
2553 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2554 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2557 struct rdma_restrack_root;
2560 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2561 struct device *dma_device;
2562 struct ib_device_ops ops;
2563 char name[IB_DEVICE_NAME_MAX];
2564 struct rcu_head rcu_head;
2566 struct list_head event_handler_list;
2567 spinlock_t event_handler_lock;
2569 struct rw_semaphore client_data_rwsem;
2570 struct xarray client_data;
2571 struct mutex unregistration_lock;
2573 struct ib_cache cache;
2575 * port_data is indexed by port number
2577 struct ib_port_data *port_data;
2579 int num_comp_vectors;
2581 struct iw_cm_verbs *iwcm;
2583 struct module *owner;
2585 /* First group for device attributes,
2586 * Second group for driver provided attributes (optional).
2587 * It is NULL terminated array.
2589 const struct attribute_group *groups[3];
2591 struct kobject *ports_kobj;
2592 struct list_head port_list;
2595 u64 uverbs_cmd_mask;
2596 u64 uverbs_ex_cmd_mask;
2598 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2602 /* Indicates kernel verbs support, should not be used in drivers */
2603 u16 kverbs_provider:1;
2606 struct ib_device_attr attrs;
2607 struct attribute_group *hw_stats_ag;
2608 struct rdma_hw_stats *hw_stats;
2610 #ifdef CONFIG_CGROUP_RDMA
2611 struct rdmacg_device cg_device;
2615 struct rdma_restrack_root *res;
2617 const struct uapi_definition *driver_def;
2618 enum rdma_driver_id driver_id;
2621 * Positive refcount indicates that the device is currently
2622 * registered and cannot be unregistered.
2624 refcount_t refcount;
2625 struct completion unreg_completion;
2626 struct work_struct unregistration_work;
2628 const struct rdma_link_ops *link_ops;
2633 void (*add) (struct ib_device *);
2634 void (*remove)(struct ib_device *, void *client_data);
2636 /* Returns the net_dev belonging to this ib_client and matching the
2638 * @dev: An RDMA device that the net_dev use for communication.
2639 * @port: A physical port number on the RDMA device.
2640 * @pkey: P_Key that the net_dev uses if applicable.
2641 * @gid: A GID that the net_dev uses to communicate.
2642 * @addr: An IP address the net_dev is configured with.
2643 * @client_data: The device's client data set by ib_set_client_data().
2645 * An ib_client that implements a net_dev on top of RDMA devices
2646 * (such as IP over IB) should implement this callback, allowing the
2647 * rdma_cm module to find the right net_dev for a given request.
2649 * The caller is responsible for calling dev_put on the returned
2651 struct net_device *(*get_net_dev_by_params)(
2652 struct ib_device *dev,
2655 const union ib_gid *gid,
2656 const struct sockaddr *addr,
2658 struct list_head list;
2661 /* kverbs are not required by the client */
2665 struct ib_device *_ib_alloc_device(size_t size);
2666 #define ib_alloc_device(drv_struct, member) \
2667 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2668 BUILD_BUG_ON_ZERO(offsetof( \
2669 struct drv_struct, member))), \
2670 struct drv_struct, member)
2672 void ib_dealloc_device(struct ib_device *device);
2674 void ib_get_device_fw_str(struct ib_device *device, char *str);
2676 int ib_register_device(struct ib_device *device, const char *name);
2677 void ib_unregister_device(struct ib_device *device);
2678 void ib_unregister_driver(enum rdma_driver_id driver_id);
2679 void ib_unregister_device_and_put(struct ib_device *device);
2680 void ib_unregister_device_queued(struct ib_device *ib_dev);
2682 int ib_register_client (struct ib_client *client);
2683 void ib_unregister_client(struct ib_client *client);
2686 * ib_get_client_data - Get IB client context
2687 * @device:Device to get context for
2688 * @client:Client to get context for
2690 * ib_get_client_data() returns the client context data set with
2691 * ib_set_client_data(). This can only be called while the client is
2692 * registered to the device, once the ib_client remove() callback returns this
2695 static inline void *ib_get_client_data(struct ib_device *device,
2696 struct ib_client *client)
2698 return xa_load(&device->client_data, client->client_id);
2700 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2702 void ib_set_device_ops(struct ib_device *device,
2703 const struct ib_device_ops *ops);
2705 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2706 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2707 unsigned long pfn, unsigned long size, pgprot_t prot);
2708 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2709 struct vm_area_struct *vma, struct page *page,
2710 unsigned long size);
2712 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2713 struct vm_area_struct *vma,
2714 unsigned long pfn, unsigned long size,
2719 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2720 struct vm_area_struct *vma, struct page *page,
2727 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2729 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2732 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2734 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2737 static inline bool ib_is_buffer_cleared(const void __user *p,
2743 if (len > USHRT_MAX)
2746 buf = memdup_user(p, len);
2750 ret = !memchr_inv(buf, 0, len);
2755 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2759 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2763 * ib_is_destroy_retryable - Check whether the uobject destruction
2765 * @ret: The initial destruction return code
2766 * @why: remove reason
2767 * @uobj: The uobject that is destroyed
2769 * This function is a helper function that IB layer and low-level drivers
2770 * can use to consider whether the destruction of the given uobject is
2772 * It checks the original return code, if it wasn't success the destruction
2773 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2774 * the remove reason. (i.e. why).
2775 * Must be called with the object locked for destroy.
2777 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2778 struct ib_uobject *uobj)
2780 return ret && (why == RDMA_REMOVE_DESTROY ||
2781 uobj->context->cleanup_retryable);
2785 * ib_destroy_usecnt - Called during destruction to check the usecnt
2786 * @usecnt: The usecnt atomic
2787 * @why: remove reason
2788 * @uobj: The uobject that is destroyed
2790 * Non-zero usecnts will block destruction unless destruction was triggered by
2791 * a ucontext cleanup.
2793 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2794 enum rdma_remove_reason why,
2795 struct ib_uobject *uobj)
2797 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2803 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2804 * contains all required attributes and no attributes not allowed for
2805 * the given QP state transition.
2806 * @cur_state: Current QP state
2807 * @next_state: Next QP state
2809 * @mask: Mask of supplied QP attributes
2811 * This function is a helper function that a low-level driver's
2812 * modify_qp method can use to validate the consumer's input. It
2813 * checks that cur_state and next_state are valid QP states, that a
2814 * transition from cur_state to next_state is allowed by the IB spec,
2815 * and that the attribute mask supplied is allowed for the transition.
2817 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2818 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2820 void ib_register_event_handler(struct ib_event_handler *event_handler);
2821 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2822 void ib_dispatch_event(struct ib_event *event);
2824 int ib_query_port(struct ib_device *device,
2825 u8 port_num, struct ib_port_attr *port_attr);
2827 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2831 * rdma_cap_ib_switch - Check if the device is IB switch
2832 * @device: Device to check
2834 * Device driver is responsible for setting is_switch bit on
2835 * in ib_device structure at init time.
2837 * Return: true if the device is IB switch.
2839 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2841 return device->is_switch;
2845 * rdma_start_port - Return the first valid port number for the device
2848 * @device: Device to be checked
2850 * Return start port number
2852 static inline u8 rdma_start_port(const struct ib_device *device)
2854 return rdma_cap_ib_switch(device) ? 0 : 1;
2858 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2859 * @device - The struct ib_device * to iterate over
2860 * @iter - The unsigned int to store the port number
2862 #define rdma_for_each_port(device, iter) \
2863 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2864 unsigned int, iter))); \
2865 iter <= rdma_end_port(device); (iter)++)
2868 * rdma_end_port - Return the last valid port number for the device
2871 * @device: Device to be checked
2873 * Return last port number
2875 static inline u8 rdma_end_port(const struct ib_device *device)
2877 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2880 static inline int rdma_is_port_valid(const struct ib_device *device,
2883 return (port >= rdma_start_port(device) &&
2884 port <= rdma_end_port(device));
2887 static inline bool rdma_is_grh_required(const struct ib_device *device,
2890 return device->port_data[port_num].immutable.core_cap_flags &
2891 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2894 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2896 return device->port_data[port_num].immutable.core_cap_flags &
2897 RDMA_CORE_CAP_PROT_IB;
2900 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2902 return device->port_data[port_num].immutable.core_cap_flags &
2903 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2906 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2908 return device->port_data[port_num].immutable.core_cap_flags &
2909 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2912 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2914 return device->port_data[port_num].immutable.core_cap_flags &
2915 RDMA_CORE_CAP_PROT_ROCE;
2918 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2920 return device->port_data[port_num].immutable.core_cap_flags &
2921 RDMA_CORE_CAP_PROT_IWARP;
2924 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2926 return rdma_protocol_ib(device, port_num) ||
2927 rdma_protocol_roce(device, port_num);
2930 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2932 return device->port_data[port_num].immutable.core_cap_flags &
2933 RDMA_CORE_CAP_PROT_RAW_PACKET;
2936 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2938 return device->port_data[port_num].immutable.core_cap_flags &
2939 RDMA_CORE_CAP_PROT_USNIC;
2943 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2944 * Management Datagrams.
2945 * @device: Device to check
2946 * @port_num: Port number to check
2948 * Management Datagrams (MAD) are a required part of the InfiniBand
2949 * specification and are supported on all InfiniBand devices. A slightly
2950 * extended version are also supported on OPA interfaces.
2952 * Return: true if the port supports sending/receiving of MAD packets.
2954 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2956 return device->port_data[port_num].immutable.core_cap_flags &
2957 RDMA_CORE_CAP_IB_MAD;
2961 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2962 * Management Datagrams.
2963 * @device: Device to check
2964 * @port_num: Port number to check
2966 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2967 * datagrams with their own versions. These OPA MADs share many but not all of
2968 * the characteristics of InfiniBand MADs.
2970 * OPA MADs differ in the following ways:
2972 * 1) MADs are variable size up to 2K
2973 * IBTA defined MADs remain fixed at 256 bytes
2974 * 2) OPA SMPs must carry valid PKeys
2975 * 3) OPA SMP packets are a different format
2977 * Return: true if the port supports OPA MAD packet formats.
2979 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2981 return (device->port_data[port_num].immutable.core_cap_flags &
2982 RDMA_CORE_CAP_OPA_MAD) == RDMA_CORE_CAP_OPA_MAD;
2986 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2987 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2988 * @device: Device to check
2989 * @port_num: Port number to check
2991 * Each InfiniBand node is required to provide a Subnet Management Agent
2992 * that the subnet manager can access. Prior to the fabric being fully
2993 * configured by the subnet manager, the SMA is accessed via a well known
2994 * interface called the Subnet Management Interface (SMI). This interface
2995 * uses directed route packets to communicate with the SM to get around the
2996 * chicken and egg problem of the SM needing to know what's on the fabric
2997 * in order to configure the fabric, and needing to configure the fabric in
2998 * order to send packets to the devices on the fabric. These directed
2999 * route packets do not need the fabric fully configured in order to reach
3000 * their destination. The SMI is the only method allowed to send
3001 * directed route packets on an InfiniBand fabric.
3003 * Return: true if the port provides an SMI.
3005 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3007 return device->port_data[port_num].immutable.core_cap_flags &
3008 RDMA_CORE_CAP_IB_SMI;
3012 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3013 * Communication Manager.
3014 * @device: Device to check
3015 * @port_num: Port number to check
3017 * The InfiniBand Communication Manager is one of many pre-defined General
3018 * Service Agents (GSA) that are accessed via the General Service
3019 * Interface (GSI). It's role is to facilitate establishment of connections
3020 * between nodes as well as other management related tasks for established
3023 * Return: true if the port supports an IB CM (this does not guarantee that
3024 * a CM is actually running however).
3026 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3028 return device->port_data[port_num].immutable.core_cap_flags &
3029 RDMA_CORE_CAP_IB_CM;
3033 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3034 * Communication Manager.
3035 * @device: Device to check
3036 * @port_num: Port number to check
3038 * Similar to above, but specific to iWARP connections which have a different
3039 * managment protocol than InfiniBand.
3041 * Return: true if the port supports an iWARP CM (this does not guarantee that
3042 * a CM is actually running however).
3044 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3046 return device->port_data[port_num].immutable.core_cap_flags &
3047 RDMA_CORE_CAP_IW_CM;
3051 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3052 * Subnet Administration.
3053 * @device: Device to check
3054 * @port_num: Port number to check
3056 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3057 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3058 * fabrics, devices should resolve routes to other hosts by contacting the
3059 * SA to query the proper route.
3061 * Return: true if the port should act as a client to the fabric Subnet
3062 * Administration interface. This does not imply that the SA service is
3065 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3067 return device->port_data[port_num].immutable.core_cap_flags &
3068 RDMA_CORE_CAP_IB_SA;
3072 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3074 * @device: Device to check
3075 * @port_num: Port number to check
3077 * InfiniBand multicast registration is more complex than normal IPv4 or
3078 * IPv6 multicast registration. Each Host Channel Adapter must register
3079 * with the Subnet Manager when it wishes to join a multicast group. It
3080 * should do so only once regardless of how many queue pairs it subscribes
3081 * to this group. And it should leave the group only after all queue pairs
3082 * attached to the group have been detached.
3084 * Return: true if the port must undertake the additional adminstrative
3085 * overhead of registering/unregistering with the SM and tracking of the
3086 * total number of queue pairs attached to the multicast group.
3088 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3090 return rdma_cap_ib_sa(device, port_num);
3094 * rdma_cap_af_ib - Check if the port of device has the capability
3095 * Native Infiniband Address.
3096 * @device: Device to check
3097 * @port_num: Port number to check
3099 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3100 * GID. RoCE uses a different mechanism, but still generates a GID via
3101 * a prescribed mechanism and port specific data.
3103 * Return: true if the port uses a GID address to identify devices on the
3106 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3108 return device->port_data[port_num].immutable.core_cap_flags &
3109 RDMA_CORE_CAP_AF_IB;
3113 * rdma_cap_eth_ah - Check if the port of device has the capability
3114 * Ethernet Address Handle.
3115 * @device: Device to check
3116 * @port_num: Port number to check
3118 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3119 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3120 * port. Normally, packet headers are generated by the sending host
3121 * adapter, but when sending connectionless datagrams, we must manually
3122 * inject the proper headers for the fabric we are communicating over.
3124 * Return: true if we are running as a RoCE port and must force the
3125 * addition of a Global Route Header built from our Ethernet Address
3126 * Handle into our header list for connectionless packets.
3128 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3130 return device->port_data[port_num].immutable.core_cap_flags &
3131 RDMA_CORE_CAP_ETH_AH;
3135 * rdma_cap_opa_ah - Check if the port of device supports
3136 * OPA Address handles
3137 * @device: Device to check
3138 * @port_num: Port number to check
3140 * Return: true if we are running on an OPA device which supports
3141 * the extended OPA addressing.
3143 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3145 return (device->port_data[port_num].immutable.core_cap_flags &
3146 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3150 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3153 * @port_num: Port number
3155 * This MAD size includes the MAD headers and MAD payload. No other headers
3158 * Return the max MAD size required by the Port. Will return 0 if the port
3159 * does not support MADs
3161 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3163 return device->port_data[port_num].immutable.max_mad_size;
3167 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3168 * @device: Device to check
3169 * @port_num: Port number to check
3171 * RoCE GID table mechanism manages the various GIDs for a device.
3173 * NOTE: if allocating the port's GID table has failed, this call will still
3174 * return true, but any RoCE GID table API will fail.
3176 * Return: true if the port uses RoCE GID table mechanism in order to manage
3179 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3182 return rdma_protocol_roce(device, port_num) &&
3183 device->ops.add_gid && device->ops.del_gid;
3187 * Check if the device supports READ W/ INVALIDATE.
3189 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3192 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3193 * has support for it yet.
3195 return rdma_protocol_iwarp(dev, port_num);
3198 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3200 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3201 struct ifla_vf_info *info);
3202 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3203 struct ifla_vf_stats *stats);
3204 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3207 int ib_query_pkey(struct ib_device *device,
3208 u8 port_num, u16 index, u16 *pkey);
3210 int ib_modify_device(struct ib_device *device,
3211 int device_modify_mask,
3212 struct ib_device_modify *device_modify);
3214 int ib_modify_port(struct ib_device *device,
3215 u8 port_num, int port_modify_mask,
3216 struct ib_port_modify *port_modify);
3218 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3219 u8 *port_num, u16 *index);
3221 int ib_find_pkey(struct ib_device *device,
3222 u8 port_num, u16 pkey, u16 *index);
3226 * Create a memory registration for all memory in the system and place
3227 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3228 * ULPs to avoid the overhead of dynamic MRs.
3230 * This flag is generally considered unsafe and must only be used in
3231 * extremly trusted environments. Every use of it will log a warning
3232 * in the kernel log.
3234 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3237 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3238 const char *caller);
3239 #define ib_alloc_pd(device, flags) \
3240 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3241 void ib_dealloc_pd(struct ib_pd *pd);
3243 enum rdma_create_ah_flags {
3244 /* In a sleepable context */
3245 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3249 * rdma_create_ah - Creates an address handle for the given address vector.
3250 * @pd: The protection domain associated with the address handle.
3251 * @ah_attr: The attributes of the address vector.
3252 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3254 * The address handle is used to reference a local or global destination
3255 * in all UD QP post sends.
3257 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3261 * rdma_create_user_ah - Creates an address handle for the given address vector.
3262 * It resolves destination mac address for ah attribute of RoCE type.
3263 * @pd: The protection domain associated with the address handle.
3264 * @ah_attr: The attributes of the address vector.
3265 * @udata: pointer to user's input output buffer information need by
3268 * It returns 0 on success and returns appropriate error code on error.
3269 * The address handle is used to reference a local or global destination
3270 * in all UD QP post sends.
3272 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3273 struct rdma_ah_attr *ah_attr,
3274 struct ib_udata *udata);
3276 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3278 * @hdr: the L3 header to parse
3279 * @net_type: type of header to parse
3280 * @sgid: place to store source gid
3281 * @dgid: place to store destination gid
3283 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3284 enum rdma_network_type net_type,
3285 union ib_gid *sgid, union ib_gid *dgid);
3288 * ib_get_rdma_header_version - Get the header version
3289 * @hdr: the L3 header to parse
3291 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3294 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3296 * @device: Device on which the received message arrived.
3297 * @port_num: Port on which the received message arrived.
3298 * @wc: Work completion associated with the received message.
3299 * @grh: References the received global route header. This parameter is
3300 * ignored unless the work completion indicates that the GRH is valid.
3301 * @ah_attr: Returned attributes that can be used when creating an address
3302 * handle for replying to the message.
3303 * When ib_init_ah_attr_from_wc() returns success,
3304 * (a) for IB link layer it optionally contains a reference to SGID attribute
3305 * when GRH is present for IB link layer.
3306 * (b) for RoCE link layer it contains a reference to SGID attribute.
3307 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3308 * attributes which are initialized using ib_init_ah_attr_from_wc().
3311 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3312 const struct ib_wc *wc, const struct ib_grh *grh,
3313 struct rdma_ah_attr *ah_attr);
3316 * ib_create_ah_from_wc - Creates an address handle associated with the
3317 * sender of the specified work completion.
3318 * @pd: The protection domain associated with the address handle.
3319 * @wc: Work completion information associated with a received message.
3320 * @grh: References the received global route header. This parameter is
3321 * ignored unless the work completion indicates that the GRH is valid.
3322 * @port_num: The outbound port number to associate with the address.
3324 * The address handle is used to reference a local or global destination
3325 * in all UD QP post sends.
3327 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3328 const struct ib_grh *grh, u8 port_num);
3331 * rdma_modify_ah - Modifies the address vector associated with an address
3333 * @ah: The address handle to modify.
3334 * @ah_attr: The new address vector attributes to associate with the
3337 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3340 * rdma_query_ah - Queries the address vector associated with an address
3342 * @ah: The address handle to query.
3343 * @ah_attr: The address vector attributes associated with the address
3346 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3348 enum rdma_destroy_ah_flags {
3349 /* In a sleepable context */
3350 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3354 * rdma_destroy_ah - Destroys an address handle.
3355 * @ah: The address handle to destroy.
3356 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3358 int rdma_destroy_ah(struct ib_ah *ah, u32 flags);
3361 * ib_create_srq - Creates a SRQ associated with the specified protection
3363 * @pd: The protection domain associated with the SRQ.
3364 * @srq_init_attr: A list of initial attributes required to create the
3365 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3366 * the actual capabilities of the created SRQ.
3368 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3369 * requested size of the SRQ, and set to the actual values allocated
3370 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3371 * will always be at least as large as the requested values.
3373 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3374 struct ib_srq_init_attr *srq_init_attr);
3377 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3378 * @srq: The SRQ to modify.
3379 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3380 * the current values of selected SRQ attributes are returned.
3381 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3382 * are being modified.
3384 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3385 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3386 * the number of receives queued drops below the limit.
3388 int ib_modify_srq(struct ib_srq *srq,
3389 struct ib_srq_attr *srq_attr,
3390 enum ib_srq_attr_mask srq_attr_mask);
3393 * ib_query_srq - Returns the attribute list and current values for the
3395 * @srq: The SRQ to query.
3396 * @srq_attr: The attributes of the specified SRQ.
3398 int ib_query_srq(struct ib_srq *srq,
3399 struct ib_srq_attr *srq_attr);
3402 * ib_destroy_srq - Destroys the specified SRQ.
3403 * @srq: The SRQ to destroy.
3405 int ib_destroy_srq(struct ib_srq *srq);
3408 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3409 * @srq: The SRQ to post the work request on.
3410 * @recv_wr: A list of work requests to post on the receive queue.
3411 * @bad_recv_wr: On an immediate failure, this parameter will reference
3412 * the work request that failed to be posted on the QP.
3414 static inline int ib_post_srq_recv(struct ib_srq *srq,
3415 const struct ib_recv_wr *recv_wr,
3416 const struct ib_recv_wr **bad_recv_wr)
3418 const struct ib_recv_wr *dummy;
3420 return srq->device->ops.post_srq_recv(srq, recv_wr,
3421 bad_recv_wr ? : &dummy);
3425 * ib_create_qp - Creates a QP associated with the specified protection
3427 * @pd: The protection domain associated with the QP.
3428 * @qp_init_attr: A list of initial attributes required to create the
3429 * QP. If QP creation succeeds, then the attributes are updated to
3430 * the actual capabilities of the created QP.
3432 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3433 struct ib_qp_init_attr *qp_init_attr);
3436 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3437 * @qp: The QP to modify.
3438 * @attr: On input, specifies the QP attributes to modify. On output,
3439 * the current values of selected QP attributes are returned.
3440 * @attr_mask: A bit-mask used to specify which attributes of the QP
3441 * are being modified.
3442 * @udata: pointer to user's input output buffer information
3443 * are being modified.
3444 * It returns 0 on success and returns appropriate error code on error.
3446 int ib_modify_qp_with_udata(struct ib_qp *qp,
3447 struct ib_qp_attr *attr,
3449 struct ib_udata *udata);
3452 * ib_modify_qp - Modifies the attributes for the specified QP and then
3453 * transitions the QP to the given state.
3454 * @qp: The QP to modify.
3455 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3456 * the current values of selected QP attributes are returned.
3457 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3458 * are being modified.
3460 int ib_modify_qp(struct ib_qp *qp,
3461 struct ib_qp_attr *qp_attr,
3465 * ib_query_qp - Returns the attribute list and current values for the
3467 * @qp: The QP to query.
3468 * @qp_attr: The attributes of the specified QP.
3469 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3470 * @qp_init_attr: Additional attributes of the selected QP.
3472 * The qp_attr_mask may be used to limit the query to gathering only the
3473 * selected attributes.
3475 int ib_query_qp(struct ib_qp *qp,
3476 struct ib_qp_attr *qp_attr,
3478 struct ib_qp_init_attr *qp_init_attr);
3481 * ib_destroy_qp - Destroys the specified QP.
3482 * @qp: The QP to destroy.
3484 int ib_destroy_qp(struct ib_qp *qp);
3487 * ib_open_qp - Obtain a reference to an existing sharable QP.
3488 * @xrcd - XRC domain
3489 * @qp_open_attr: Attributes identifying the QP to open.
3491 * Returns a reference to a sharable QP.
3493 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3494 struct ib_qp_open_attr *qp_open_attr);
3497 * ib_close_qp - Release an external reference to a QP.
3498 * @qp: The QP handle to release
3500 * The opened QP handle is released by the caller. The underlying
3501 * shared QP is not destroyed until all internal references are released.
3503 int ib_close_qp(struct ib_qp *qp);
3506 * ib_post_send - Posts a list of work requests to the send queue of
3508 * @qp: The QP to post the work request on.
3509 * @send_wr: A list of work requests to post on the send queue.
3510 * @bad_send_wr: On an immediate failure, this parameter will reference
3511 * the work request that failed to be posted on the QP.
3513 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3514 * error is returned, the QP state shall not be affected,
3515 * ib_post_send() will return an immediate error after queueing any
3516 * earlier work requests in the list.
3518 static inline int ib_post_send(struct ib_qp *qp,
3519 const struct ib_send_wr *send_wr,
3520 const struct ib_send_wr **bad_send_wr)
3522 const struct ib_send_wr *dummy;
3524 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3528 * ib_post_recv - Posts a list of work requests to the receive queue of
3530 * @qp: The QP to post the work request on.
3531 * @recv_wr: A list of work requests to post on the receive queue.
3532 * @bad_recv_wr: On an immediate failure, this parameter will reference
3533 * the work request that failed to be posted on the QP.
3535 static inline int ib_post_recv(struct ib_qp *qp,
3536 const struct ib_recv_wr *recv_wr,
3537 const struct ib_recv_wr **bad_recv_wr)
3539 const struct ib_recv_wr *dummy;
3541 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3544 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3545 int nr_cqe, int comp_vector,
3546 enum ib_poll_context poll_ctx, const char *caller);
3547 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3548 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3550 void ib_free_cq(struct ib_cq *cq);
3551 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3554 * ib_create_cq - Creates a CQ on the specified device.
3555 * @device: The device on which to create the CQ.
3556 * @comp_handler: A user-specified callback that is invoked when a
3557 * completion event occurs on the CQ.
3558 * @event_handler: A user-specified callback that is invoked when an
3559 * asynchronous event not associated with a completion occurs on the CQ.
3560 * @cq_context: Context associated with the CQ returned to the user via
3561 * the associated completion and event handlers.
3562 * @cq_attr: The attributes the CQ should be created upon.
3564 * Users can examine the cq structure to determine the actual CQ size.
3566 struct ib_cq *__ib_create_cq(struct ib_device *device,
3567 ib_comp_handler comp_handler,
3568 void (*event_handler)(struct ib_event *, void *),
3570 const struct ib_cq_init_attr *cq_attr,
3571 const char *caller);
3572 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3573 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3576 * ib_resize_cq - Modifies the capacity of the CQ.
3577 * @cq: The CQ to resize.
3578 * @cqe: The minimum size of the CQ.
3580 * Users can examine the cq structure to determine the actual CQ size.
3582 int ib_resize_cq(struct ib_cq *cq, int cqe);
3585 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3586 * @cq: The CQ to modify.
3587 * @cq_count: number of CQEs that will trigger an event
3588 * @cq_period: max period of time in usec before triggering an event
3591 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3594 * ib_destroy_cq - Destroys the specified CQ.
3595 * @cq: The CQ to destroy.
3597 int ib_destroy_cq(struct ib_cq *cq);
3600 * ib_poll_cq - poll a CQ for completion(s)
3601 * @cq:the CQ being polled
3602 * @num_entries:maximum number of completions to return
3603 * @wc:array of at least @num_entries &struct ib_wc where completions
3606 * Poll a CQ for (possibly multiple) completions. If the return value
3607 * is < 0, an error occurred. If the return value is >= 0, it is the
3608 * number of completions returned. If the return value is
3609 * non-negative and < num_entries, then the CQ was emptied.
3611 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3614 return cq->device->ops.poll_cq(cq, num_entries, wc);
3618 * ib_req_notify_cq - Request completion notification on a CQ.
3619 * @cq: The CQ to generate an event for.
3621 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3622 * to request an event on the next solicited event or next work
3623 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3624 * may also be |ed in to request a hint about missed events, as
3628 * < 0 means an error occurred while requesting notification
3629 * == 0 means notification was requested successfully, and if
3630 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3631 * were missed and it is safe to wait for another event. In
3632 * this case is it guaranteed that any work completions added
3633 * to the CQ since the last CQ poll will trigger a completion
3634 * notification event.
3635 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3636 * in. It means that the consumer must poll the CQ again to
3637 * make sure it is empty to avoid missing an event because of a
3638 * race between requesting notification and an entry being
3639 * added to the CQ. This return value means it is possible
3640 * (but not guaranteed) that a work completion has been added
3641 * to the CQ since the last poll without triggering a
3642 * completion notification event.
3644 static inline int ib_req_notify_cq(struct ib_cq *cq,
3645 enum ib_cq_notify_flags flags)
3647 return cq->device->ops.req_notify_cq(cq, flags);
3651 * ib_req_ncomp_notif - Request completion notification when there are
3652 * at least the specified number of unreaped completions on the CQ.
3653 * @cq: The CQ to generate an event for.
3654 * @wc_cnt: The number of unreaped completions that should be on the
3655 * CQ before an event is generated.
3657 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3659 return cq->device->ops.req_ncomp_notif ?
3660 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3665 * ib_dma_mapping_error - check a DMA addr for error
3666 * @dev: The device for which the dma_addr was created
3667 * @dma_addr: The DMA address to check
3669 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3671 return dma_mapping_error(dev->dma_device, dma_addr);
3675 * ib_dma_map_single - Map a kernel virtual address to DMA address
3676 * @dev: The device for which the dma_addr is to be created
3677 * @cpu_addr: The kernel virtual address
3678 * @size: The size of the region in bytes
3679 * @direction: The direction of the DMA
3681 static inline u64 ib_dma_map_single(struct ib_device *dev,
3682 void *cpu_addr, size_t size,
3683 enum dma_data_direction direction)
3685 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3689 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3690 * @dev: The device for which the DMA address was created
3691 * @addr: The DMA address
3692 * @size: The size of the region in bytes
3693 * @direction: The direction of the DMA
3695 static inline void ib_dma_unmap_single(struct ib_device *dev,
3696 u64 addr, size_t size,
3697 enum dma_data_direction direction)
3699 dma_unmap_single(dev->dma_device, addr, size, direction);
3703 * ib_dma_map_page - Map a physical page to DMA address
3704 * @dev: The device for which the dma_addr is to be created
3705 * @page: The page to be mapped
3706 * @offset: The offset within the page
3707 * @size: The size of the region in bytes
3708 * @direction: The direction of the DMA
3710 static inline u64 ib_dma_map_page(struct ib_device *dev,
3712 unsigned long offset,
3714 enum dma_data_direction direction)
3716 return dma_map_page(dev->dma_device, page, offset, size, direction);
3720 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3721 * @dev: The device for which the DMA address was created
3722 * @addr: The DMA address
3723 * @size: The size of the region in bytes
3724 * @direction: The direction of the DMA
3726 static inline void ib_dma_unmap_page(struct ib_device *dev,
3727 u64 addr, size_t size,
3728 enum dma_data_direction direction)
3730 dma_unmap_page(dev->dma_device, addr, size, direction);
3734 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3735 * @dev: The device for which the DMA addresses are to be created
3736 * @sg: The array of scatter/gather entries
3737 * @nents: The number of scatter/gather entries
3738 * @direction: The direction of the DMA
3740 static inline int ib_dma_map_sg(struct ib_device *dev,
3741 struct scatterlist *sg, int nents,
3742 enum dma_data_direction direction)
3744 return dma_map_sg(dev->dma_device, sg, nents, direction);
3748 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3749 * @dev: The device for which the DMA addresses were created
3750 * @sg: The array of scatter/gather entries
3751 * @nents: The number of scatter/gather entries
3752 * @direction: The direction of the DMA
3754 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3755 struct scatterlist *sg, int nents,
3756 enum dma_data_direction direction)
3758 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3761 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3762 struct scatterlist *sg, int nents,
3763 enum dma_data_direction direction,
3764 unsigned long dma_attrs)
3766 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3770 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3771 struct scatterlist *sg, int nents,
3772 enum dma_data_direction direction,
3773 unsigned long dma_attrs)
3775 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3779 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3780 * @dev: The device to query
3782 * The returned value represents a size in bytes.
3784 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3786 struct device_dma_parameters *p = dev->dma_device->dma_parms;
3788 return p ? p->max_segment_size : UINT_MAX;
3792 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3793 * @dev: The device for which the DMA address was created
3794 * @addr: The DMA address
3795 * @size: The size of the region in bytes
3796 * @dir: The direction of the DMA
3798 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3801 enum dma_data_direction dir)
3803 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3807 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3808 * @dev: The device for which the DMA address was created
3809 * @addr: The DMA address
3810 * @size: The size of the region in bytes
3811 * @dir: The direction of the DMA
3813 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3816 enum dma_data_direction dir)
3818 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3822 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3823 * @dev: The device for which the DMA address is requested
3824 * @size: The size of the region to allocate in bytes
3825 * @dma_handle: A pointer for returning the DMA address of the region
3826 * @flag: memory allocator flags
3828 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3830 dma_addr_t *dma_handle,
3833 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3837 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3838 * @dev: The device for which the DMA addresses were allocated
3839 * @size: The size of the region
3840 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3841 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3843 static inline void ib_dma_free_coherent(struct ib_device *dev,
3844 size_t size, void *cpu_addr,
3845 dma_addr_t dma_handle)
3847 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3851 * ib_dereg_mr - Deregisters a memory region and removes it from the
3852 * HCA translation table.
3853 * @mr: The memory region to deregister.
3855 * This function can fail, if the memory region has memory windows bound to it.
3857 int ib_dereg_mr(struct ib_mr *mr);
3859 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3860 enum ib_mr_type mr_type,
3864 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3866 * @mr - struct ib_mr pointer to be updated.
3867 * @newkey - new key to be used.
3869 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3871 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3872 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3876 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3877 * for calculating a new rkey for type 2 memory windows.
3878 * @rkey - the rkey to increment.
3880 static inline u32 ib_inc_rkey(u32 rkey)
3882 const u32 mask = 0x000000ff;
3883 return ((rkey + 1) & mask) | (rkey & ~mask);
3887 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3888 * @pd: The protection domain associated with the unmapped region.
3889 * @mr_access_flags: Specifies the memory access rights.
3890 * @fmr_attr: Attributes of the unmapped region.
3892 * A fast memory region must be mapped before it can be used as part of
3895 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3896 int mr_access_flags,
3897 struct ib_fmr_attr *fmr_attr);
3900 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3901 * @fmr: The fast memory region to associate with the pages.
3902 * @page_list: An array of physical pages to map to the fast memory region.
3903 * @list_len: The number of pages in page_list.
3904 * @iova: The I/O virtual address to use with the mapped region.
3906 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3907 u64 *page_list, int list_len,
3910 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
3914 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3915 * @fmr_list: A linked list of fast memory regions to unmap.
3917 int ib_unmap_fmr(struct list_head *fmr_list);
3920 * ib_dealloc_fmr - Deallocates a fast memory region.
3921 * @fmr: The fast memory region to deallocate.
3923 int ib_dealloc_fmr(struct ib_fmr *fmr);
3926 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3927 * @qp: QP to attach to the multicast group. The QP must be type
3929 * @gid: Multicast group GID.
3930 * @lid: Multicast group LID in host byte order.
3932 * In order to send and receive multicast packets, subnet
3933 * administration must have created the multicast group and configured
3934 * the fabric appropriately. The port associated with the specified
3935 * QP must also be a member of the multicast group.
3937 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3940 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3941 * @qp: QP to detach from the multicast group.
3942 * @gid: Multicast group GID.
3943 * @lid: Multicast group LID in host byte order.
3945 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3948 * ib_alloc_xrcd - Allocates an XRC domain.
3949 * @device: The device on which to allocate the XRC domain.
3950 * @caller: Module name for kernel consumers
3952 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3953 #define ib_alloc_xrcd(device) \
3954 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3957 * ib_dealloc_xrcd - Deallocates an XRC domain.
3958 * @xrcd: The XRC domain to deallocate.
3960 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3962 static inline int ib_check_mr_access(int flags)
3965 * Local write permission is required if remote write or
3966 * remote atomic permission is also requested.
3968 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3969 !(flags & IB_ACCESS_LOCAL_WRITE))
3975 static inline bool ib_access_writable(int access_flags)
3978 * We have writable memory backing the MR if any of the following
3979 * access flags are set. "Local write" and "remote write" obviously
3980 * require write access. "Remote atomic" can do things like fetch and
3981 * add, which will modify memory, and "MW bind" can change permissions
3982 * by binding a window.
3984 return access_flags &
3985 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3986 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3990 * ib_check_mr_status: lightweight check of MR status.
3991 * This routine may provide status checks on a selected
3992 * ib_mr. first use is for signature status check.
3994 * @mr: A memory region.
3995 * @check_mask: Bitmask of which checks to perform from
3996 * ib_mr_status_check enumeration.
3997 * @mr_status: The container of relevant status checks.
3998 * failed checks will be indicated in the status bitmask
3999 * and the relevant info shall be in the error item.
4001 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4002 struct ib_mr_status *mr_status);
4005 * ib_device_try_get: Hold a registration lock
4006 * device: The device to lock
4008 * A device under an active registration lock cannot become unregistered. It
4009 * is only possible to obtain a registration lock on a device that is fully
4010 * registered, otherwise this function returns false.
4012 * The registration lock is only necessary for actions which require the
4013 * device to still be registered. Uses that only require the device pointer to
4014 * be valid should use get_device(&ibdev->dev) to hold the memory.
4017 static inline bool ib_device_try_get(struct ib_device *dev)
4019 return refcount_inc_not_zero(&dev->refcount);
4022 void ib_device_put(struct ib_device *device);
4023 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4024 enum rdma_driver_id driver_id);
4025 struct ib_device *ib_device_get_by_name(const char *name,
4026 enum rdma_driver_id driver_id);
4027 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4028 u16 pkey, const union ib_gid *gid,
4029 const struct sockaddr *addr);
4030 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4032 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4034 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4035 struct ib_wq_init_attr *init_attr);
4036 int ib_destroy_wq(struct ib_wq *wq);
4037 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4039 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4040 struct ib_rwq_ind_table_init_attr*
4041 wq_ind_table_init_attr);
4042 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4044 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4045 unsigned int *sg_offset, unsigned int page_size);
4048 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4049 unsigned int *sg_offset, unsigned int page_size)
4053 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4059 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4060 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4062 void ib_drain_rq(struct ib_qp *qp);
4063 void ib_drain_sq(struct ib_qp *qp);
4064 void ib_drain_qp(struct ib_qp *qp);
4066 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4068 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4070 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4071 return attr->roce.dmac;
4075 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4077 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4078 attr->ib.dlid = (u16)dlid;
4079 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4080 attr->opa.dlid = dlid;
4083 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4085 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4086 return attr->ib.dlid;
4087 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4088 return attr->opa.dlid;
4092 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4097 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4102 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4105 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4106 attr->ib.src_path_bits = src_path_bits;
4107 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4108 attr->opa.src_path_bits = src_path_bits;
4111 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4113 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4114 return attr->ib.src_path_bits;
4115 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4116 return attr->opa.src_path_bits;
4120 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4123 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4124 attr->opa.make_grd = make_grd;
4127 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4129 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4130 return attr->opa.make_grd;
4134 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4136 attr->port_num = port_num;
4139 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4141 return attr->port_num;
4144 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4147 attr->static_rate = static_rate;
4150 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4152 return attr->static_rate;
4155 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4156 enum ib_ah_flags flag)
4158 attr->ah_flags = flag;
4161 static inline enum ib_ah_flags
4162 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4164 return attr->ah_flags;
4167 static inline const struct ib_global_route
4168 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4173 /*To retrieve and modify the grh */
4174 static inline struct ib_global_route
4175 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4180 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4182 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4184 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4187 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4190 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4192 grh->dgid.global.subnet_prefix = prefix;
4195 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4198 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4200 grh->dgid.global.interface_id = if_id;
4203 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4204 union ib_gid *dgid, u32 flow_label,
4205 u8 sgid_index, u8 hop_limit,
4208 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4210 attr->ah_flags = IB_AH_GRH;
4213 grh->flow_label = flow_label;
4214 grh->sgid_index = sgid_index;
4215 grh->hop_limit = hop_limit;
4216 grh->traffic_class = traffic_class;
4217 grh->sgid_attr = NULL;
4220 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4221 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4222 u32 flow_label, u8 hop_limit, u8 traffic_class,
4223 const struct ib_gid_attr *sgid_attr);
4224 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4225 const struct rdma_ah_attr *src);
4226 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4227 const struct rdma_ah_attr *new);
4228 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4231 * rdma_ah_find_type - Return address handle type.
4233 * @dev: Device to be checked
4234 * @port_num: Port number
4236 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4239 if (rdma_protocol_roce(dev, port_num))
4240 return RDMA_AH_ATTR_TYPE_ROCE;
4241 if (rdma_protocol_ib(dev, port_num)) {
4242 if (rdma_cap_opa_ah(dev, port_num))
4243 return RDMA_AH_ATTR_TYPE_OPA;
4244 return RDMA_AH_ATTR_TYPE_IB;
4247 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4251 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4252 * In the current implementation the only way to get
4253 * get the 32bit lid is from other sources for OPA.
4254 * For IB, lids will always be 16bits so cast the
4255 * value accordingly.
4259 static inline u16 ib_lid_cpu16(u32 lid)
4261 WARN_ON_ONCE(lid & 0xFFFF0000);
4266 * ib_lid_be16 - Return lid in 16bit BE encoding.
4270 static inline __be16 ib_lid_be16(u32 lid)
4272 WARN_ON_ONCE(lid & 0xFFFF0000);
4273 return cpu_to_be16((u16)lid);
4277 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4279 * @device: the rdma device
4280 * @comp_vector: index of completion vector
4282 * Returns NULL on failure, otherwise a corresponding cpu map of the
4283 * completion vector (returns all-cpus map if the device driver doesn't
4284 * implement get_vector_affinity).
4286 static inline const struct cpumask *
4287 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4289 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4290 !device->ops.get_vector_affinity)
4293 return device->ops.get_vector_affinity(device, comp_vector);
4298 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4299 * and add their gids, as needed, to the relevant RoCE devices.
4301 * @device: the rdma device
4303 void rdma_roce_rescan_device(struct ib_device *ibdev);
4305 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4307 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4309 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4310 enum rdma_netdev_t type, const char *name,
4311 unsigned char name_assign_type,
4312 void (*setup)(struct net_device *));
4314 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4315 enum rdma_netdev_t type, const char *name,
4316 unsigned char name_assign_type,
4317 void (*setup)(struct net_device *),
4318 struct net_device *netdev);
4321 * rdma_set_device_sysfs_group - Set device attributes group to have
4322 * driver specific sysfs entries at
4323 * for infiniband class.
4325 * @device: device pointer for which attributes to be created
4326 * @group: Pointer to group which should be added when device
4327 * is registered with sysfs.
4328 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4329 * group per device to have sysfs attributes.
4331 * NOTE: New drivers should not make use of this API; instead new device
4332 * parameter should be exposed via netlink command. This API and mechanism
4333 * exist only for existing drivers.
4336 rdma_set_device_sysfs_group(struct ib_device *dev,
4337 const struct attribute_group *group)
4339 dev->groups[1] = group;
4343 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4345 * @device: device pointer for which ib_device pointer to retrieve
4347 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4350 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4352 return container_of(device, struct ib_device, dev);
4356 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4357 * ib_device holder structure from device pointer.
4359 * NOTE: New drivers should not make use of this API; This API is only for
4360 * existing drivers who have exposed sysfs entries using
4361 * rdma_set_device_sysfs_group().
4363 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4364 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4365 #endif /* IB_VERBS_H */