2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
67 #include <uapi/rdma/rdma_user_ioctl.h>
68 #include <uapi/rdma/ib_user_ioctl_verbs.h>
70 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
74 extern struct workqueue_struct *ib_wq;
75 extern struct workqueue_struct *ib_comp_wq;
76 extern struct workqueue_struct *ib_comp_unbound_wq;
86 extern union ib_gid zgid;
89 /* If link layer is Ethernet, this is RoCE V1 */
92 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
96 #define ROCE_V2_UDP_DPORT 4791
98 struct net_device *ndev;
99 struct ib_device *device;
101 enum ib_gid_type gid_type;
106 enum rdma_node_type {
107 /* IB values map to NodeInfo:NodeType. */
117 /* set the local administered indication */
118 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
121 enum rdma_transport_type {
123 RDMA_TRANSPORT_IWARP,
124 RDMA_TRANSPORT_USNIC,
125 RDMA_TRANSPORT_USNIC_UDP
128 enum rdma_protocol_type {
132 RDMA_PROTOCOL_USNIC_UDP
135 __attribute_const__ enum rdma_transport_type
136 rdma_node_get_transport(enum rdma_node_type node_type);
138 enum rdma_network_type {
140 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
145 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
147 if (network_type == RDMA_NETWORK_IPV4 ||
148 network_type == RDMA_NETWORK_IPV6)
149 return IB_GID_TYPE_ROCE_UDP_ENCAP;
151 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
152 return IB_GID_TYPE_IB;
155 static inline enum rdma_network_type
156 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
158 if (attr->gid_type == IB_GID_TYPE_IB)
159 return RDMA_NETWORK_IB;
161 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
162 return RDMA_NETWORK_IPV4;
164 return RDMA_NETWORK_IPV6;
167 enum rdma_link_layer {
168 IB_LINK_LAYER_UNSPECIFIED,
169 IB_LINK_LAYER_INFINIBAND,
170 IB_LINK_LAYER_ETHERNET,
173 enum ib_device_cap_flags {
174 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
175 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
176 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
177 IB_DEVICE_RAW_MULTI = (1 << 3),
178 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
179 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
180 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
181 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
182 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
183 /* Not in use, former INIT_TYPE = (1 << 9),*/
184 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
185 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
186 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
187 IB_DEVICE_SRQ_RESIZE = (1 << 13),
188 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
191 * This device supports a per-device lkey or stag that can be
192 * used without performing a memory registration for the local
193 * memory. Note that ULPs should never check this flag, but
194 * instead of use the local_dma_lkey flag in the ib_pd structure,
195 * which will always contain a usable lkey.
197 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
198 /* Reserved, old SEND_W_INV = (1 << 16),*/
199 IB_DEVICE_MEM_WINDOW = (1 << 17),
201 * Devices should set IB_DEVICE_UD_IP_SUM if they support
202 * insertion of UDP and TCP checksum on outgoing UD IPoIB
203 * messages and can verify the validity of checksum for
204 * incoming messages. Setting this flag implies that the
205 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
207 IB_DEVICE_UD_IP_CSUM = (1 << 18),
208 IB_DEVICE_UD_TSO = (1 << 19),
209 IB_DEVICE_XRC = (1 << 20),
212 * This device supports the IB "base memory management extension",
213 * which includes support for fast registrations (IB_WR_REG_MR,
214 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
215 * also be set by any iWarp device which must support FRs to comply
216 * to the iWarp verbs spec. iWarp devices also support the
217 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
220 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
221 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
222 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
223 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
224 IB_DEVICE_RC_IP_CSUM = (1 << 25),
225 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
226 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
228 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
229 * support execution of WQEs that involve synchronization
230 * of I/O operations with single completion queue managed
233 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
234 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
235 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
236 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
237 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
238 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
239 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
240 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
241 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
242 /* The device supports padding incoming writes to cacheline. */
243 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
246 enum ib_signature_prot_cap {
247 IB_PROT_T10DIF_TYPE_1 = 1,
248 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
249 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
252 enum ib_signature_guard_cap {
253 IB_GUARD_T10DIF_CRC = 1,
254 IB_GUARD_T10DIF_CSUM = 1 << 1,
263 enum ib_odp_general_cap_bits {
264 IB_ODP_SUPPORT = 1 << 0,
265 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
268 enum ib_odp_transport_cap_bits {
269 IB_ODP_SUPPORT_SEND = 1 << 0,
270 IB_ODP_SUPPORT_RECV = 1 << 1,
271 IB_ODP_SUPPORT_WRITE = 1 << 2,
272 IB_ODP_SUPPORT_READ = 1 << 3,
273 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
277 uint64_t general_caps;
279 uint32_t rc_odp_caps;
280 uint32_t uc_odp_caps;
281 uint32_t ud_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
445 static inline int ib_width_enum_to_int(enum ib_port_width width)
448 case IB_WIDTH_1X: return 1;
449 case IB_WIDTH_4X: return 4;
450 case IB_WIDTH_8X: return 8;
451 case IB_WIDTH_12X: return 12;
467 * struct rdma_hw_stats
468 * @lock - Mutex to protect parallel write access to lifespan and values
469 * of counters, which are 64bits and not guaranteeed to be written
470 * atomicaly on 32bits systems.
471 * @timestamp - Used by the core code to track when the last update was
472 * @lifespan - Used by the core code to determine how old the counters
473 * should be before being updated again. Stored in jiffies, defaults
474 * to 10 milliseconds, drivers can override the default be specifying
475 * their own value during their allocation routine.
476 * @name - Array of pointers to static names used for the counters in
478 * @num_counters - How many hardware counters there are. If name is
479 * shorter than this number, a kernel oops will result. Driver authors
480 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
481 * in their code to prevent this.
482 * @value - Array of u64 counters that are accessed by the sysfs code and
483 * filled in by the drivers get_stats routine
485 struct rdma_hw_stats {
486 struct mutex lock; /* Protect lifespan and values[] */
487 unsigned long timestamp;
488 unsigned long lifespan;
489 const char * const *names;
494 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
496 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
498 * @names - Array of static const char *
499 * @num_counters - How many elements in array
500 * @lifespan - How many milliseconds between updates
502 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
503 const char * const *names, int num_counters,
504 unsigned long lifespan)
506 struct rdma_hw_stats *stats;
508 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
512 stats->names = names;
513 stats->num_counters = num_counters;
514 stats->lifespan = msecs_to_jiffies(lifespan);
520 /* Define bits for the various functionality this port needs to be supported by
523 /* Management 0x00000FFF */
524 #define RDMA_CORE_CAP_IB_MAD 0x00000001
525 #define RDMA_CORE_CAP_IB_SMI 0x00000002
526 #define RDMA_CORE_CAP_IB_CM 0x00000004
527 #define RDMA_CORE_CAP_IW_CM 0x00000008
528 #define RDMA_CORE_CAP_IB_SA 0x00000010
529 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
531 /* Address format 0x000FF000 */
532 #define RDMA_CORE_CAP_AF_IB 0x00001000
533 #define RDMA_CORE_CAP_ETH_AH 0x00002000
534 #define RDMA_CORE_CAP_OPA_AH 0x00004000
535 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
537 /* Protocol 0xFFF00000 */
538 #define RDMA_CORE_CAP_PROT_IB 0x00100000
539 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
540 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
541 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
542 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
543 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
545 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
546 | RDMA_CORE_CAP_PROT_ROCE \
547 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
549 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
550 | RDMA_CORE_CAP_IB_MAD \
551 | RDMA_CORE_CAP_IB_SMI \
552 | RDMA_CORE_CAP_IB_CM \
553 | RDMA_CORE_CAP_IB_SA \
554 | RDMA_CORE_CAP_AF_IB)
555 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
556 | RDMA_CORE_CAP_IB_MAD \
557 | RDMA_CORE_CAP_IB_CM \
558 | RDMA_CORE_CAP_AF_IB \
559 | RDMA_CORE_CAP_ETH_AH)
560 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
561 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
562 | RDMA_CORE_CAP_IB_MAD \
563 | RDMA_CORE_CAP_IB_CM \
564 | RDMA_CORE_CAP_AF_IB \
565 | RDMA_CORE_CAP_ETH_AH)
566 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
567 | RDMA_CORE_CAP_IW_CM)
568 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
569 | RDMA_CORE_CAP_OPA_MAD)
571 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
573 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
575 struct ib_port_attr {
577 enum ib_port_state state;
579 enum ib_mtu active_mtu;
581 unsigned int ip_gids:1;
582 /* This is the value from PortInfo CapabilityMask, defined by IBA */
600 enum ib_device_modify_flags {
601 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
602 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
605 #define IB_DEVICE_NODE_DESC_MAX 64
607 struct ib_device_modify {
609 char node_desc[IB_DEVICE_NODE_DESC_MAX];
612 enum ib_port_modify_flags {
613 IB_PORT_SHUTDOWN = 1,
614 IB_PORT_INIT_TYPE = (1<<2),
615 IB_PORT_RESET_QKEY_CNTR = (1<<3),
616 IB_PORT_OPA_MASK_CHG = (1<<4)
619 struct ib_port_modify {
620 u32 set_port_cap_mask;
621 u32 clr_port_cap_mask;
629 IB_EVENT_QP_ACCESS_ERR,
633 IB_EVENT_PATH_MIG_ERR,
634 IB_EVENT_DEVICE_FATAL,
635 IB_EVENT_PORT_ACTIVE,
638 IB_EVENT_PKEY_CHANGE,
641 IB_EVENT_SRQ_LIMIT_REACHED,
642 IB_EVENT_QP_LAST_WQE_REACHED,
643 IB_EVENT_CLIENT_REREGISTER,
648 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
651 struct ib_device *device;
659 enum ib_event_type event;
662 struct ib_event_handler {
663 struct ib_device *device;
664 void (*handler)(struct ib_event_handler *, struct ib_event *);
665 struct list_head list;
668 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
670 (_ptr)->device = _device; \
671 (_ptr)->handler = _handler; \
672 INIT_LIST_HEAD(&(_ptr)->list); \
675 struct ib_global_route {
676 const struct ib_gid_attr *sgid_attr;
685 __be32 version_tclass_flow;
693 union rdma_network_hdr {
696 /* The IB spec states that if it's IPv4, the header
697 * is located in the last 20 bytes of the header.
700 struct iphdr roce4grh;
704 #define IB_QPN_MASK 0xFFFFFF
707 IB_MULTICAST_QPN = 0xffffff
710 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
711 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
718 IB_RATE_PORT_CURRENT = 0,
719 IB_RATE_2_5_GBPS = 2,
727 IB_RATE_120_GBPS = 10,
728 IB_RATE_14_GBPS = 11,
729 IB_RATE_56_GBPS = 12,
730 IB_RATE_112_GBPS = 13,
731 IB_RATE_168_GBPS = 14,
732 IB_RATE_25_GBPS = 15,
733 IB_RATE_100_GBPS = 16,
734 IB_RATE_200_GBPS = 17,
735 IB_RATE_300_GBPS = 18
739 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
740 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
741 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
742 * @rate: rate to convert.
744 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
747 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
748 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
749 * @rate: rate to convert.
751 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
755 * enum ib_mr_type - memory region type
756 * @IB_MR_TYPE_MEM_REG: memory region that is used for
757 * normal registration
758 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
759 * signature operations (data-integrity
761 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
762 * register any arbitrary sg lists (without
763 * the normal mr constraints - see
768 IB_MR_TYPE_SIGNATURE,
774 * IB_SIG_TYPE_NONE: Unprotected.
775 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
777 enum ib_signature_type {
783 * Signature T10-DIF block-guard types
784 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
785 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
787 enum ib_t10_dif_bg_type {
793 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
795 * @bg_type: T10-DIF block guard type (CRC|CSUM)
796 * @pi_interval: protection information interval.
797 * @bg: seed of guard computation.
798 * @app_tag: application tag of guard block
799 * @ref_tag: initial guard block reference tag.
800 * @ref_remap: Indicate wethear the reftag increments each block
801 * @app_escape: Indicate to skip block check if apptag=0xffff
802 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
803 * @apptag_check_mask: check bitmask of application tag.
805 struct ib_t10_dif_domain {
806 enum ib_t10_dif_bg_type bg_type;
814 u16 apptag_check_mask;
818 * struct ib_sig_domain - Parameters for signature domain
819 * @sig_type: specific signauture type
820 * @sig: union of all signature domain attributes that may
821 * be used to set domain layout.
823 struct ib_sig_domain {
824 enum ib_signature_type sig_type;
826 struct ib_t10_dif_domain dif;
831 * struct ib_sig_attrs - Parameters for signature handover operation
832 * @check_mask: bitmask for signature byte check (8 bytes)
833 * @mem: memory domain layout desciptor.
834 * @wire: wire domain layout desciptor.
836 struct ib_sig_attrs {
838 struct ib_sig_domain mem;
839 struct ib_sig_domain wire;
842 enum ib_sig_err_type {
849 * Signature check masks (8 bytes in total) according to the T10-PI standard:
850 * -------- -------- ------------
851 * | GUARD | APPTAG | REFTAG |
853 * -------- -------- ------------
856 IB_SIG_CHECK_GUARD = 0xc0,
857 IB_SIG_CHECK_APPTAG = 0x30,
858 IB_SIG_CHECK_REFTAG = 0x0f,
862 * struct ib_sig_err - signature error descriptor
865 enum ib_sig_err_type err_type;
872 enum ib_mr_status_check {
873 IB_MR_CHECK_SIG_STATUS = 1,
877 * struct ib_mr_status - Memory region status container
879 * @fail_status: Bitmask of MR checks status. For each
880 * failed check a corresponding status bit is set.
881 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
884 struct ib_mr_status {
886 struct ib_sig_err sig_err;
890 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
892 * @mult: multiple to convert.
894 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
896 enum rdma_ah_attr_type {
897 RDMA_AH_ATTR_TYPE_UNDEFINED,
898 RDMA_AH_ATTR_TYPE_IB,
899 RDMA_AH_ATTR_TYPE_ROCE,
900 RDMA_AH_ATTR_TYPE_OPA,
908 struct roce_ah_attr {
918 struct rdma_ah_attr {
919 struct ib_global_route grh;
924 enum rdma_ah_attr_type type;
926 struct ib_ah_attr ib;
927 struct roce_ah_attr roce;
928 struct opa_ah_attr opa;
936 IB_WC_LOC_EEC_OP_ERR,
941 IB_WC_LOC_ACCESS_ERR,
942 IB_WC_REM_INV_REQ_ERR,
943 IB_WC_REM_ACCESS_ERR,
946 IB_WC_RNR_RETRY_EXC_ERR,
947 IB_WC_LOC_RDD_VIOL_ERR,
948 IB_WC_REM_INV_RD_REQ_ERR,
951 IB_WC_INV_EEC_STATE_ERR,
953 IB_WC_RESP_TIMEOUT_ERR,
957 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
968 IB_WC_MASKED_COMP_SWAP,
969 IB_WC_MASKED_FETCH_ADD,
971 * Set value of IB_WC_RECV so consumers can test if a completion is a
972 * receive by testing (opcode & IB_WC_RECV).
975 IB_WC_RECV_RDMA_WITH_IMM
980 IB_WC_WITH_IMM = (1<<1),
981 IB_WC_WITH_INVALIDATE = (1<<2),
982 IB_WC_IP_CSUM_OK = (1<<3),
983 IB_WC_WITH_SMAC = (1<<4),
984 IB_WC_WITH_VLAN = (1<<5),
985 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
991 struct ib_cqe *wr_cqe;
993 enum ib_wc_status status;
994 enum ib_wc_opcode opcode;
1000 u32 invalidate_rkey;
1008 u8 port_num; /* valid only for DR SMPs on switches */
1011 u8 network_hdr_type;
1014 enum ib_cq_notify_flags {
1015 IB_CQ_SOLICITED = 1 << 0,
1016 IB_CQ_NEXT_COMP = 1 << 1,
1017 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1018 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1027 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1029 return srq_type == IB_SRQT_XRC ||
1030 srq_type == IB_SRQT_TM;
1033 enum ib_srq_attr_mask {
1034 IB_SRQ_MAX_WR = 1 << 0,
1035 IB_SRQ_LIMIT = 1 << 1,
1038 struct ib_srq_attr {
1044 struct ib_srq_init_attr {
1045 void (*event_handler)(struct ib_event *, void *);
1047 struct ib_srq_attr attr;
1048 enum ib_srq_type srq_type;
1054 struct ib_xrcd *xrcd;
1069 u32 max_inline_data;
1072 * Maximum number of rdma_rw_ctx structures in flight at a time.
1073 * ib_create_qp() will calculate the right amount of neededed WRs
1074 * and MRs based on this.
1086 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1087 * here (and in that order) since the MAD layer uses them as
1088 * indices into a 2-entry table.
1097 IB_QPT_RAW_ETHERTYPE,
1098 IB_QPT_RAW_PACKET = 8,
1102 IB_QPT_DRIVER = 0xFF,
1103 /* Reserve a range for qp types internal to the low level driver.
1104 * These qp types will not be visible at the IB core layer, so the
1105 * IB_QPT_MAX usages should not be affected in the core layer
1107 IB_QPT_RESERVED1 = 0x1000,
1119 enum ib_qp_create_flags {
1120 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1121 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1122 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1123 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1124 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1125 IB_QP_CREATE_NETIF_QP = 1 << 5,
1126 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1127 /* FREE = 1 << 7, */
1128 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1129 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1130 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1131 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1132 /* reserve bits 26-31 for low level drivers' internal use */
1133 IB_QP_CREATE_RESERVED_START = 1 << 26,
1134 IB_QP_CREATE_RESERVED_END = 1 << 31,
1138 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1139 * callback to destroy the passed in QP.
1142 struct ib_qp_init_attr {
1143 /* Consumer's event_handler callback must not block */
1144 void (*event_handler)(struct ib_event *, void *);
1147 struct ib_cq *send_cq;
1148 struct ib_cq *recv_cq;
1150 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1151 struct ib_qp_cap cap;
1152 enum ib_sig_type sq_sig_type;
1153 enum ib_qp_type qp_type;
1157 * Only needed for special QP types, or when using the RW API.
1160 struct ib_rwq_ind_table *rwq_ind_tbl;
1164 struct ib_qp_open_attr {
1165 void (*event_handler)(struct ib_event *, void *);
1168 enum ib_qp_type qp_type;
1171 enum ib_rnr_timeout {
1172 IB_RNR_TIMER_655_36 = 0,
1173 IB_RNR_TIMER_000_01 = 1,
1174 IB_RNR_TIMER_000_02 = 2,
1175 IB_RNR_TIMER_000_03 = 3,
1176 IB_RNR_TIMER_000_04 = 4,
1177 IB_RNR_TIMER_000_06 = 5,
1178 IB_RNR_TIMER_000_08 = 6,
1179 IB_RNR_TIMER_000_12 = 7,
1180 IB_RNR_TIMER_000_16 = 8,
1181 IB_RNR_TIMER_000_24 = 9,
1182 IB_RNR_TIMER_000_32 = 10,
1183 IB_RNR_TIMER_000_48 = 11,
1184 IB_RNR_TIMER_000_64 = 12,
1185 IB_RNR_TIMER_000_96 = 13,
1186 IB_RNR_TIMER_001_28 = 14,
1187 IB_RNR_TIMER_001_92 = 15,
1188 IB_RNR_TIMER_002_56 = 16,
1189 IB_RNR_TIMER_003_84 = 17,
1190 IB_RNR_TIMER_005_12 = 18,
1191 IB_RNR_TIMER_007_68 = 19,
1192 IB_RNR_TIMER_010_24 = 20,
1193 IB_RNR_TIMER_015_36 = 21,
1194 IB_RNR_TIMER_020_48 = 22,
1195 IB_RNR_TIMER_030_72 = 23,
1196 IB_RNR_TIMER_040_96 = 24,
1197 IB_RNR_TIMER_061_44 = 25,
1198 IB_RNR_TIMER_081_92 = 26,
1199 IB_RNR_TIMER_122_88 = 27,
1200 IB_RNR_TIMER_163_84 = 28,
1201 IB_RNR_TIMER_245_76 = 29,
1202 IB_RNR_TIMER_327_68 = 30,
1203 IB_RNR_TIMER_491_52 = 31
1206 enum ib_qp_attr_mask {
1208 IB_QP_CUR_STATE = (1<<1),
1209 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1210 IB_QP_ACCESS_FLAGS = (1<<3),
1211 IB_QP_PKEY_INDEX = (1<<4),
1212 IB_QP_PORT = (1<<5),
1213 IB_QP_QKEY = (1<<6),
1215 IB_QP_PATH_MTU = (1<<8),
1216 IB_QP_TIMEOUT = (1<<9),
1217 IB_QP_RETRY_CNT = (1<<10),
1218 IB_QP_RNR_RETRY = (1<<11),
1219 IB_QP_RQ_PSN = (1<<12),
1220 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1221 IB_QP_ALT_PATH = (1<<14),
1222 IB_QP_MIN_RNR_TIMER = (1<<15),
1223 IB_QP_SQ_PSN = (1<<16),
1224 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1225 IB_QP_PATH_MIG_STATE = (1<<18),
1226 IB_QP_CAP = (1<<19),
1227 IB_QP_DEST_QPN = (1<<20),
1228 IB_QP_RESERVED1 = (1<<21),
1229 IB_QP_RESERVED2 = (1<<22),
1230 IB_QP_RESERVED3 = (1<<23),
1231 IB_QP_RESERVED4 = (1<<24),
1232 IB_QP_RATE_LIMIT = (1<<25),
1257 enum ib_qp_state qp_state;
1258 enum ib_qp_state cur_qp_state;
1259 enum ib_mtu path_mtu;
1260 enum ib_mig_state path_mig_state;
1265 int qp_access_flags;
1266 struct ib_qp_cap cap;
1267 struct rdma_ah_attr ah_attr;
1268 struct rdma_ah_attr alt_ah_attr;
1271 u8 en_sqd_async_notify;
1274 u8 max_dest_rd_atomic;
1286 /* These are shared with userspace */
1287 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1288 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1289 IB_WR_SEND = IB_UVERBS_WR_SEND,
1290 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1291 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1292 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1293 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1294 IB_WR_LSO = IB_UVERBS_WR_TSO,
1295 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1296 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1297 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1298 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1299 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1300 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1301 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1303 /* These are kernel only and can not be issued by userspace */
1304 IB_WR_REG_MR = 0x20,
1307 /* reserve values for low level drivers' internal use.
1308 * These values will not be used at all in the ib core layer.
1310 IB_WR_RESERVED1 = 0xf0,
1322 enum ib_send_flags {
1324 IB_SEND_SIGNALED = (1<<1),
1325 IB_SEND_SOLICITED = (1<<2),
1326 IB_SEND_INLINE = (1<<3),
1327 IB_SEND_IP_CSUM = (1<<4),
1329 /* reserve bits 26-31 for low level drivers' internal use */
1330 IB_SEND_RESERVED_START = (1 << 26),
1331 IB_SEND_RESERVED_END = (1 << 31),
1341 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1345 struct ib_send_wr *next;
1348 struct ib_cqe *wr_cqe;
1350 struct ib_sge *sg_list;
1352 enum ib_wr_opcode opcode;
1356 u32 invalidate_rkey;
1361 struct ib_send_wr wr;
1366 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1368 return container_of(wr, struct ib_rdma_wr, wr);
1371 struct ib_atomic_wr {
1372 struct ib_send_wr wr;
1376 u64 compare_add_mask;
1381 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1383 return container_of(wr, struct ib_atomic_wr, wr);
1387 struct ib_send_wr wr;
1394 u16 pkey_index; /* valid for GSI only */
1395 u8 port_num; /* valid for DR SMPs on switch only */
1398 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1400 return container_of(wr, struct ib_ud_wr, wr);
1404 struct ib_send_wr wr;
1410 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1412 return container_of(wr, struct ib_reg_wr, wr);
1415 struct ib_sig_handover_wr {
1416 struct ib_send_wr wr;
1417 struct ib_sig_attrs *sig_attrs;
1418 struct ib_mr *sig_mr;
1420 struct ib_sge *prot;
1423 static inline const struct ib_sig_handover_wr *
1424 sig_handover_wr(const struct ib_send_wr *wr)
1426 return container_of(wr, struct ib_sig_handover_wr, wr);
1430 struct ib_recv_wr *next;
1433 struct ib_cqe *wr_cqe;
1435 struct ib_sge *sg_list;
1439 enum ib_access_flags {
1440 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1441 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1442 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1443 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1444 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1445 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1446 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1447 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1449 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1453 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1454 * are hidden here instead of a uapi header!
1456 enum ib_mr_rereg_flags {
1457 IB_MR_REREG_TRANS = 1,
1458 IB_MR_REREG_PD = (1<<1),
1459 IB_MR_REREG_ACCESS = (1<<2),
1460 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1463 struct ib_fmr_attr {
1471 enum rdma_remove_reason {
1473 * Userspace requested uobject deletion or initial try
1474 * to remove uobject via cleanup. Call could fail
1476 RDMA_REMOVE_DESTROY,
1477 /* Context deletion. This call should delete the actual object itself */
1479 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1480 RDMA_REMOVE_DRIVER_REMOVE,
1481 /* uobj is being cleaned-up before being committed */
1485 struct ib_rdmacg_object {
1486 #ifdef CONFIG_CGROUP_RDMA
1487 struct rdma_cgroup *cg; /* owner rdma cgroup */
1491 struct ib_ucontext {
1492 struct ib_device *device;
1493 struct ib_uverbs_file *ufile;
1495 * 'closing' can be read by the driver only during a destroy callback,
1496 * it is set when we are closing the file descriptor and indicates
1497 * that mm_sem may be locked.
1501 bool cleanup_retryable;
1503 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1504 void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1505 unsigned long start, unsigned long end);
1506 struct mutex per_mm_list_lock;
1507 struct list_head per_mm_list;
1510 struct ib_rdmacg_object cg_obj;
1514 u64 user_handle; /* handle given to us by userspace */
1515 /* ufile & ucontext owning this object */
1516 struct ib_uverbs_file *ufile;
1517 /* FIXME, save memory: ufile->context == context */
1518 struct ib_ucontext *context; /* associated user context */
1519 void *object; /* containing object */
1520 struct list_head list; /* link to context's list */
1521 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1522 int id; /* index into kernel idr */
1524 atomic_t usecnt; /* protects exclusive access */
1525 struct rcu_head rcu; /* kfree_rcu() overhead */
1527 const struct uverbs_api_object *uapi_object;
1531 const void __user *inbuf;
1532 void __user *outbuf;
1540 struct ib_device *device;
1541 struct ib_uobject *uobject;
1542 atomic_t usecnt; /* count all resources */
1544 u32 unsafe_global_rkey;
1547 * Implementation details of the RDMA core, don't use in drivers:
1549 struct ib_mr *__internal_mr;
1550 struct rdma_restrack_entry res;
1554 struct ib_device *device;
1555 atomic_t usecnt; /* count all exposed resources */
1556 struct inode *inode;
1558 struct mutex tgt_qp_mutex;
1559 struct list_head tgt_qp_list;
1563 struct ib_device *device;
1565 struct ib_uobject *uobject;
1566 const struct ib_gid_attr *sgid_attr;
1567 enum rdma_ah_attr_type type;
1570 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1572 enum ib_poll_context {
1573 IB_POLL_DIRECT, /* caller context, no hw completions */
1574 IB_POLL_SOFTIRQ, /* poll from softirq context */
1575 IB_POLL_WORKQUEUE, /* poll from workqueue */
1576 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1580 struct ib_device *device;
1581 struct ib_uobject *uobject;
1582 ib_comp_handler comp_handler;
1583 void (*event_handler)(struct ib_event *, void *);
1586 atomic_t usecnt; /* count number of work queues */
1587 enum ib_poll_context poll_ctx;
1590 struct irq_poll iop;
1591 struct work_struct work;
1593 struct workqueue_struct *comp_wq;
1595 * Implementation details of the RDMA core, don't use in drivers:
1597 struct rdma_restrack_entry res;
1601 struct ib_device *device;
1603 struct ib_uobject *uobject;
1604 void (*event_handler)(struct ib_event *, void *);
1606 enum ib_srq_type srq_type;
1613 struct ib_xrcd *xrcd;
1620 enum ib_raw_packet_caps {
1621 /* Strip cvlan from incoming packet and report it in the matching work
1622 * completion is supported.
1624 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1625 /* Scatter FCS field of an incoming packet to host memory is supported.
1627 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1628 /* Checksum offloads are supported (for both send and receive). */
1629 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1630 /* When a packet is received for an RQ with no receive WQEs, the
1631 * packet processing is delayed.
1633 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1647 struct ib_device *device;
1648 struct ib_uobject *uobject;
1650 void (*event_handler)(struct ib_event *, void *);
1654 enum ib_wq_state state;
1655 enum ib_wq_type wq_type;
1660 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1661 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1662 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1663 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1666 struct ib_wq_init_attr {
1668 enum ib_wq_type wq_type;
1672 void (*event_handler)(struct ib_event *, void *);
1673 u32 create_flags; /* Use enum ib_wq_flags */
1676 enum ib_wq_attr_mask {
1677 IB_WQ_STATE = 1 << 0,
1678 IB_WQ_CUR_STATE = 1 << 1,
1679 IB_WQ_FLAGS = 1 << 2,
1683 enum ib_wq_state wq_state;
1684 enum ib_wq_state curr_wq_state;
1685 u32 flags; /* Use enum ib_wq_flags */
1686 u32 flags_mask; /* Use enum ib_wq_flags */
1689 struct ib_rwq_ind_table {
1690 struct ib_device *device;
1691 struct ib_uobject *uobject;
1694 u32 log_ind_tbl_size;
1695 struct ib_wq **ind_tbl;
1698 struct ib_rwq_ind_table_init_attr {
1699 u32 log_ind_tbl_size;
1700 /* Each entry is a pointer to Receive Work Queue */
1701 struct ib_wq **ind_tbl;
1704 enum port_pkey_state {
1705 IB_PORT_PKEY_NOT_VALID = 0,
1706 IB_PORT_PKEY_VALID = 1,
1707 IB_PORT_PKEY_LISTED = 2,
1710 struct ib_qp_security;
1712 struct ib_port_pkey {
1713 enum port_pkey_state state;
1716 struct list_head qp_list;
1717 struct list_head to_error_list;
1718 struct ib_qp_security *sec;
1721 struct ib_ports_pkeys {
1722 struct ib_port_pkey main;
1723 struct ib_port_pkey alt;
1726 struct ib_qp_security {
1728 struct ib_device *dev;
1729 /* Hold this mutex when changing port and pkey settings. */
1731 struct ib_ports_pkeys *ports_pkeys;
1732 /* A list of all open shared QP handles. Required to enforce security
1733 * properly for all users of a shared QP.
1735 struct list_head shared_qp_list;
1738 atomic_t error_list_count;
1739 struct completion error_complete;
1740 int error_comps_pending;
1744 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1745 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1748 struct ib_device *device;
1750 struct ib_cq *send_cq;
1751 struct ib_cq *recv_cq;
1754 struct list_head rdma_mrs;
1755 struct list_head sig_mrs;
1757 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1758 struct list_head xrcd_list;
1760 /* count times opened, mcast attaches, flow attaches */
1762 struct list_head open_list;
1763 struct ib_qp *real_qp;
1764 struct ib_uobject *uobject;
1765 void (*event_handler)(struct ib_event *, void *);
1767 /* sgid_attrs associated with the AV's */
1768 const struct ib_gid_attr *av_sgid_attr;
1769 const struct ib_gid_attr *alt_path_sgid_attr;
1773 enum ib_qp_type qp_type;
1774 struct ib_rwq_ind_table *rwq_ind_tbl;
1775 struct ib_qp_security *qp_sec;
1779 * Implementation details of the RDMA core, don't use in drivers:
1781 struct rdma_restrack_entry res;
1785 struct ib_device *device;
1788 struct ib_uobject *uobject;
1793 struct ib_device *device;
1799 unsigned int page_size;
1802 struct ib_uobject *uobject; /* user */
1803 struct list_head qp_entry; /* FR */
1809 * Implementation details of the RDMA core, don't use in drivers:
1811 struct rdma_restrack_entry res;
1815 struct ib_device *device;
1817 struct ib_uobject *uobject;
1819 enum ib_mw_type type;
1823 struct ib_device *device;
1825 struct list_head list;
1830 /* Supported steering options */
1831 enum ib_flow_attr_type {
1832 /* steering according to rule specifications */
1833 IB_FLOW_ATTR_NORMAL = 0x0,
1834 /* default unicast and multicast rule -
1835 * receive all Eth traffic which isn't steered to any QP
1837 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1838 /* default multicast rule -
1839 * receive all Eth multicast traffic which isn't steered to any QP
1841 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1842 /* sniffer rule - receive all port traffic */
1843 IB_FLOW_ATTR_SNIFFER = 0x3
1846 /* Supported steering header types */
1847 enum ib_flow_spec_type {
1849 IB_FLOW_SPEC_ETH = 0x20,
1850 IB_FLOW_SPEC_IB = 0x22,
1852 IB_FLOW_SPEC_IPV4 = 0x30,
1853 IB_FLOW_SPEC_IPV6 = 0x31,
1854 IB_FLOW_SPEC_ESP = 0x34,
1856 IB_FLOW_SPEC_TCP = 0x40,
1857 IB_FLOW_SPEC_UDP = 0x41,
1858 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1859 IB_FLOW_SPEC_GRE = 0x51,
1860 IB_FLOW_SPEC_MPLS = 0x60,
1861 IB_FLOW_SPEC_INNER = 0x100,
1863 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1864 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1865 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1866 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1868 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1869 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1871 /* Flow steering rule priority is set according to it's domain.
1872 * Lower domain value means higher priority.
1874 enum ib_flow_domain {
1875 IB_FLOW_DOMAIN_USER,
1876 IB_FLOW_DOMAIN_ETHTOOL,
1879 IB_FLOW_DOMAIN_NUM /* Must be last */
1882 enum ib_flow_flags {
1883 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1884 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1885 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1888 struct ib_flow_eth_filter {
1897 struct ib_flow_spec_eth {
1900 struct ib_flow_eth_filter val;
1901 struct ib_flow_eth_filter mask;
1904 struct ib_flow_ib_filter {
1911 struct ib_flow_spec_ib {
1914 struct ib_flow_ib_filter val;
1915 struct ib_flow_ib_filter mask;
1918 /* IPv4 header flags */
1919 enum ib_ipv4_flags {
1920 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1921 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1922 last have this flag set */
1925 struct ib_flow_ipv4_filter {
1936 struct ib_flow_spec_ipv4 {
1939 struct ib_flow_ipv4_filter val;
1940 struct ib_flow_ipv4_filter mask;
1943 struct ib_flow_ipv6_filter {
1954 struct ib_flow_spec_ipv6 {
1957 struct ib_flow_ipv6_filter val;
1958 struct ib_flow_ipv6_filter mask;
1961 struct ib_flow_tcp_udp_filter {
1968 struct ib_flow_spec_tcp_udp {
1971 struct ib_flow_tcp_udp_filter val;
1972 struct ib_flow_tcp_udp_filter mask;
1975 struct ib_flow_tunnel_filter {
1980 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1981 * the tunnel_id from val has the vni value
1983 struct ib_flow_spec_tunnel {
1986 struct ib_flow_tunnel_filter val;
1987 struct ib_flow_tunnel_filter mask;
1990 struct ib_flow_esp_filter {
1997 struct ib_flow_spec_esp {
2000 struct ib_flow_esp_filter val;
2001 struct ib_flow_esp_filter mask;
2004 struct ib_flow_gre_filter {
2005 __be16 c_ks_res0_ver;
2012 struct ib_flow_spec_gre {
2015 struct ib_flow_gre_filter val;
2016 struct ib_flow_gre_filter mask;
2019 struct ib_flow_mpls_filter {
2025 struct ib_flow_spec_mpls {
2028 struct ib_flow_mpls_filter val;
2029 struct ib_flow_mpls_filter mask;
2032 struct ib_flow_spec_action_tag {
2033 enum ib_flow_spec_type type;
2038 struct ib_flow_spec_action_drop {
2039 enum ib_flow_spec_type type;
2043 struct ib_flow_spec_action_handle {
2044 enum ib_flow_spec_type type;
2046 struct ib_flow_action *act;
2049 enum ib_counters_description {
2054 struct ib_flow_spec_action_count {
2055 enum ib_flow_spec_type type;
2057 struct ib_counters *counters;
2060 union ib_flow_spec {
2065 struct ib_flow_spec_eth eth;
2066 struct ib_flow_spec_ib ib;
2067 struct ib_flow_spec_ipv4 ipv4;
2068 struct ib_flow_spec_tcp_udp tcp_udp;
2069 struct ib_flow_spec_ipv6 ipv6;
2070 struct ib_flow_spec_tunnel tunnel;
2071 struct ib_flow_spec_esp esp;
2072 struct ib_flow_spec_gre gre;
2073 struct ib_flow_spec_mpls mpls;
2074 struct ib_flow_spec_action_tag flow_tag;
2075 struct ib_flow_spec_action_drop drop;
2076 struct ib_flow_spec_action_handle action;
2077 struct ib_flow_spec_action_count flow_count;
2080 struct ib_flow_attr {
2081 enum ib_flow_attr_type type;
2087 union ib_flow_spec flows[];
2092 struct ib_device *device;
2093 struct ib_uobject *uobject;
2096 enum ib_flow_action_type {
2097 IB_FLOW_ACTION_UNSPECIFIED,
2098 IB_FLOW_ACTION_ESP = 1,
2101 struct ib_flow_action_attrs_esp_keymats {
2102 enum ib_uverbs_flow_action_esp_keymat protocol;
2104 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2108 struct ib_flow_action_attrs_esp_replays {
2109 enum ib_uverbs_flow_action_esp_replay protocol;
2111 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2115 enum ib_flow_action_attrs_esp_flags {
2116 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2117 * This is done in order to share the same flags between user-space and
2118 * kernel and spare an unnecessary translation.
2122 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2123 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2126 struct ib_flow_spec_list {
2127 struct ib_flow_spec_list *next;
2128 union ib_flow_spec spec;
2131 struct ib_flow_action_attrs_esp {
2132 struct ib_flow_action_attrs_esp_keymats *keymat;
2133 struct ib_flow_action_attrs_esp_replays *replay;
2134 struct ib_flow_spec_list *encap;
2135 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2136 * Value of 0 is a valid value.
2142 /* Use enum ib_flow_action_attrs_esp_flags */
2144 u64 hard_limit_pkts;
2147 struct ib_flow_action {
2148 struct ib_device *device;
2149 struct ib_uobject *uobject;
2150 enum ib_flow_action_type type;
2157 enum ib_process_mad_flags {
2158 IB_MAD_IGNORE_MKEY = 1,
2159 IB_MAD_IGNORE_BKEY = 2,
2160 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2163 enum ib_mad_result {
2164 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2165 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2166 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2167 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2170 struct ib_port_cache {
2172 struct ib_pkey_cache *pkey;
2173 struct ib_gid_table *gid;
2175 enum ib_port_state port_state;
2180 struct ib_event_handler event_handler;
2181 struct ib_port_cache *ports;
2186 struct ib_port_immutable {
2193 /* rdma netdev type - specifies protocol type */
2194 enum rdma_netdev_t {
2195 RDMA_NETDEV_OPA_VNIC,
2200 * struct rdma_netdev - rdma netdev
2201 * For cases where netstack interfacing is required.
2203 struct rdma_netdev {
2205 struct ib_device *hca;
2209 * cleanup function must be specified.
2210 * FIXME: This is only used for OPA_VNIC and that usage should be
2213 void (*free_rdma_netdev)(struct net_device *netdev);
2215 /* control functions */
2216 void (*set_id)(struct net_device *netdev, int id);
2218 int (*send)(struct net_device *dev, struct sk_buff *skb,
2219 struct ib_ah *address, u32 dqpn);
2221 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2222 union ib_gid *gid, u16 mlid,
2223 int set_qkey, u32 qkey);
2224 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2225 union ib_gid *gid, u16 mlid);
2228 struct rdma_netdev_alloc_params {
2234 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2235 struct net_device *netdev, void *param);
2238 struct ib_port_pkey_list {
2239 /* Lock to hold while modifying the list. */
2240 spinlock_t list_lock;
2241 struct list_head pkey_list;
2244 struct ib_counters {
2245 struct ib_device *device;
2246 struct ib_uobject *uobject;
2247 /* num of objects attached */
2251 struct ib_counters_read_attr {
2254 u32 flags; /* use enum ib_read_counters_flags */
2257 struct uverbs_attr_bundle;
2260 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2261 struct device *dma_device;
2263 char name[IB_DEVICE_NAME_MAX];
2265 struct list_head event_handler_list;
2266 spinlock_t event_handler_lock;
2268 rwlock_t client_data_lock;
2269 struct list_head core_list;
2270 /* Access to the client_data_list is protected by the client_data_lock
2271 * rwlock and the lists_rwsem read-write semaphore
2273 struct list_head client_data_list;
2275 struct ib_cache cache;
2277 * port_immutable is indexed by port number
2279 struct ib_port_immutable *port_immutable;
2281 int num_comp_vectors;
2283 struct ib_port_pkey_list *port_pkey_list;
2285 struct iw_cm_verbs *iwcm;
2288 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2289 * driver initialized data. The struct is kfree()'ed by the sysfs
2290 * core when the device is removed. A lifespan of -1 in the return
2291 * struct tells the core to set a default lifespan.
2293 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2296 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2297 * @index - The index in the value array we wish to have updated, or
2298 * num_counters if we want all stats updated
2300 * < 0 - Error, no counters updated
2301 * index - Updated the single counter pointed to by index
2302 * num_counters - Updated all counters (will reset the timestamp
2303 * and prevent further calls for lifespan milliseconds)
2304 * Drivers are allowed to update all counters in leiu of just the
2305 * one given in index at their option
2307 int (*get_hw_stats)(struct ib_device *device,
2308 struct rdma_hw_stats *stats,
2309 u8 port, int index);
2310 int (*query_device)(struct ib_device *device,
2311 struct ib_device_attr *device_attr,
2312 struct ib_udata *udata);
2313 int (*query_port)(struct ib_device *device,
2315 struct ib_port_attr *port_attr);
2316 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2318 /* When calling get_netdev, the HW vendor's driver should return the
2319 * net device of device @device at port @port_num or NULL if such
2320 * a net device doesn't exist. The vendor driver should call dev_hold
2321 * on this net device. The HW vendor's device driver must guarantee
2322 * that this function returns NULL before the net device has finished
2323 * NETDEV_UNREGISTER state.
2325 struct net_device *(*get_netdev)(struct ib_device *device,
2327 /* query_gid should be return GID value for @device, when @port_num
2328 * link layer is either IB or iWarp. It is no-op if @port_num port
2329 * is RoCE link layer.
2331 int (*query_gid)(struct ib_device *device,
2332 u8 port_num, int index,
2334 /* When calling add_gid, the HW vendor's driver should add the gid
2335 * of device of port at gid index available at @attr. Meta-info of
2336 * that gid (for example, the network device related to this gid) is
2337 * available at @attr. @context allows the HW vendor driver to store
2338 * extra information together with a GID entry. The HW vendor driver may
2339 * allocate memory to contain this information and store it in @context
2340 * when a new GID entry is written to. Params are consistent until the
2341 * next call of add_gid or delete_gid. The function should return 0 on
2342 * success or error otherwise. The function could be called
2343 * concurrently for different ports. This function is only called when
2344 * roce_gid_table is used.
2346 int (*add_gid)(const struct ib_gid_attr *attr,
2348 /* When calling del_gid, the HW vendor's driver should delete the
2349 * gid of device @device at gid index gid_index of port port_num
2350 * available in @attr.
2351 * Upon the deletion of a GID entry, the HW vendor must free any
2352 * allocated memory. The caller will clear @context afterwards.
2353 * This function is only called when roce_gid_table is used.
2355 int (*del_gid)(const struct ib_gid_attr *attr,
2357 int (*query_pkey)(struct ib_device *device,
2358 u8 port_num, u16 index, u16 *pkey);
2359 int (*modify_device)(struct ib_device *device,
2360 int device_modify_mask,
2361 struct ib_device_modify *device_modify);
2362 int (*modify_port)(struct ib_device *device,
2363 u8 port_num, int port_modify_mask,
2364 struct ib_port_modify *port_modify);
2365 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2366 struct ib_udata *udata);
2367 int (*dealloc_ucontext)(struct ib_ucontext *context);
2368 int (*mmap)(struct ib_ucontext *context,
2369 struct vm_area_struct *vma);
2370 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2371 struct ib_ucontext *context,
2372 struct ib_udata *udata);
2373 int (*dealloc_pd)(struct ib_pd *pd);
2374 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2375 struct rdma_ah_attr *ah_attr,
2376 struct ib_udata *udata);
2377 int (*modify_ah)(struct ib_ah *ah,
2378 struct rdma_ah_attr *ah_attr);
2379 int (*query_ah)(struct ib_ah *ah,
2380 struct rdma_ah_attr *ah_attr);
2381 int (*destroy_ah)(struct ib_ah *ah);
2382 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2383 struct ib_srq_init_attr *srq_init_attr,
2384 struct ib_udata *udata);
2385 int (*modify_srq)(struct ib_srq *srq,
2386 struct ib_srq_attr *srq_attr,
2387 enum ib_srq_attr_mask srq_attr_mask,
2388 struct ib_udata *udata);
2389 int (*query_srq)(struct ib_srq *srq,
2390 struct ib_srq_attr *srq_attr);
2391 int (*destroy_srq)(struct ib_srq *srq);
2392 int (*post_srq_recv)(struct ib_srq *srq,
2393 const struct ib_recv_wr *recv_wr,
2394 const struct ib_recv_wr **bad_recv_wr);
2395 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2396 struct ib_qp_init_attr *qp_init_attr,
2397 struct ib_udata *udata);
2398 int (*modify_qp)(struct ib_qp *qp,
2399 struct ib_qp_attr *qp_attr,
2401 struct ib_udata *udata);
2402 int (*query_qp)(struct ib_qp *qp,
2403 struct ib_qp_attr *qp_attr,
2405 struct ib_qp_init_attr *qp_init_attr);
2406 int (*destroy_qp)(struct ib_qp *qp);
2407 int (*post_send)(struct ib_qp *qp,
2408 const struct ib_send_wr *send_wr,
2409 const struct ib_send_wr **bad_send_wr);
2410 int (*post_recv)(struct ib_qp *qp,
2411 const struct ib_recv_wr *recv_wr,
2412 const struct ib_recv_wr **bad_recv_wr);
2413 struct ib_cq * (*create_cq)(struct ib_device *device,
2414 const struct ib_cq_init_attr *attr,
2415 struct ib_ucontext *context,
2416 struct ib_udata *udata);
2417 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2419 int (*destroy_cq)(struct ib_cq *cq);
2420 int (*resize_cq)(struct ib_cq *cq, int cqe,
2421 struct ib_udata *udata);
2422 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2424 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2425 int (*req_notify_cq)(struct ib_cq *cq,
2426 enum ib_cq_notify_flags flags);
2427 int (*req_ncomp_notif)(struct ib_cq *cq,
2429 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2430 int mr_access_flags);
2431 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2432 u64 start, u64 length,
2434 int mr_access_flags,
2435 struct ib_udata *udata);
2436 int (*rereg_user_mr)(struct ib_mr *mr,
2438 u64 start, u64 length,
2440 int mr_access_flags,
2442 struct ib_udata *udata);
2443 int (*dereg_mr)(struct ib_mr *mr);
2444 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2445 enum ib_mr_type mr_type,
2447 int (*map_mr_sg)(struct ib_mr *mr,
2448 struct scatterlist *sg,
2450 unsigned int *sg_offset);
2451 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2452 enum ib_mw_type type,
2453 struct ib_udata *udata);
2454 int (*dealloc_mw)(struct ib_mw *mw);
2455 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2456 int mr_access_flags,
2457 struct ib_fmr_attr *fmr_attr);
2458 int (*map_phys_fmr)(struct ib_fmr *fmr,
2459 u64 *page_list, int list_len,
2461 int (*unmap_fmr)(struct list_head *fmr_list);
2462 int (*dealloc_fmr)(struct ib_fmr *fmr);
2463 int (*attach_mcast)(struct ib_qp *qp,
2466 int (*detach_mcast)(struct ib_qp *qp,
2469 int (*process_mad)(struct ib_device *device,
2470 int process_mad_flags,
2472 const struct ib_wc *in_wc,
2473 const struct ib_grh *in_grh,
2474 const struct ib_mad_hdr *in_mad,
2476 struct ib_mad_hdr *out_mad,
2477 size_t *out_mad_size,
2478 u16 *out_mad_pkey_index);
2479 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2480 struct ib_ucontext *ucontext,
2481 struct ib_udata *udata);
2482 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2483 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2487 struct ib_udata *udata);
2488 int (*destroy_flow)(struct ib_flow *flow_id);
2489 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2490 struct ib_mr_status *mr_status);
2491 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2492 void (*drain_rq)(struct ib_qp *qp);
2493 void (*drain_sq)(struct ib_qp *qp);
2494 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2496 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2497 struct ifla_vf_info *ivf);
2498 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2499 struct ifla_vf_stats *stats);
2500 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2502 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2503 struct ib_wq_init_attr *init_attr,
2504 struct ib_udata *udata);
2505 int (*destroy_wq)(struct ib_wq *wq);
2506 int (*modify_wq)(struct ib_wq *wq,
2507 struct ib_wq_attr *attr,
2509 struct ib_udata *udata);
2510 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2511 struct ib_rwq_ind_table_init_attr *init_attr,
2512 struct ib_udata *udata);
2513 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2514 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2515 const struct ib_flow_action_attrs_esp *attr,
2516 struct uverbs_attr_bundle *attrs);
2517 int (*destroy_flow_action)(struct ib_flow_action *action);
2518 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2519 const struct ib_flow_action_attrs_esp *attr,
2520 struct uverbs_attr_bundle *attrs);
2521 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2522 struct ib_ucontext *context,
2523 struct ib_dm_alloc_attr *attr,
2524 struct uverbs_attr_bundle *attrs);
2525 int (*dealloc_dm)(struct ib_dm *dm);
2526 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2527 struct ib_dm_mr_attr *attr,
2528 struct uverbs_attr_bundle *attrs);
2529 struct ib_counters * (*create_counters)(struct ib_device *device,
2530 struct uverbs_attr_bundle *attrs);
2531 int (*destroy_counters)(struct ib_counters *counters);
2532 int (*read_counters)(struct ib_counters *counters,
2533 struct ib_counters_read_attr *counters_read_attr,
2534 struct uverbs_attr_bundle *attrs);
2537 * rdma netdev operation
2539 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2540 * must return -EOPNOTSUPP if it doesn't support the specified type.
2542 struct net_device *(*alloc_rdma_netdev)(
2543 struct ib_device *device,
2545 enum rdma_netdev_t type,
2547 unsigned char name_assign_type,
2548 void (*setup)(struct net_device *));
2550 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2551 enum rdma_netdev_t type,
2552 struct rdma_netdev_alloc_params *params);
2554 struct module *owner;
2556 /* First group for device attributes,
2557 * Second group for driver provided attributes (optional).
2558 * It is NULL terminated array.
2560 const struct attribute_group *groups[3];
2562 struct kobject *ports_kobj;
2563 struct list_head port_list;
2566 IB_DEV_UNINITIALIZED,
2572 u64 uverbs_cmd_mask;
2573 u64 uverbs_ex_cmd_mask;
2575 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2581 struct ib_device_attr attrs;
2582 struct attribute_group *hw_stats_ag;
2583 struct rdma_hw_stats *hw_stats;
2585 #ifdef CONFIG_CGROUP_RDMA
2586 struct rdmacg_device cg_device;
2591 * Implementation details of the RDMA core, don't use in drivers
2593 struct rdma_restrack_root res;
2596 * The following mandatory functions are used only at device
2597 * registration. Keep functions such as these at the end of this
2598 * structure to avoid cache line misses when accessing struct ib_device
2601 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2602 void (*get_dev_fw_str)(struct ib_device *, char *str);
2603 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2606 const struct uverbs_object_tree_def *const *driver_specs;
2607 enum rdma_driver_id driver_id;
2612 void (*add) (struct ib_device *);
2613 void (*remove)(struct ib_device *, void *client_data);
2615 /* Returns the net_dev belonging to this ib_client and matching the
2617 * @dev: An RDMA device that the net_dev use for communication.
2618 * @port: A physical port number on the RDMA device.
2619 * @pkey: P_Key that the net_dev uses if applicable.
2620 * @gid: A GID that the net_dev uses to communicate.
2621 * @addr: An IP address the net_dev is configured with.
2622 * @client_data: The device's client data set by ib_set_client_data().
2624 * An ib_client that implements a net_dev on top of RDMA devices
2625 * (such as IP over IB) should implement this callback, allowing the
2626 * rdma_cm module to find the right net_dev for a given request.
2628 * The caller is responsible for calling dev_put on the returned
2630 struct net_device *(*get_net_dev_by_params)(
2631 struct ib_device *dev,
2634 const union ib_gid *gid,
2635 const struct sockaddr *addr,
2637 struct list_head list;
2640 struct ib_device *ib_alloc_device(size_t size);
2641 void ib_dealloc_device(struct ib_device *device);
2643 void ib_get_device_fw_str(struct ib_device *device, char *str);
2645 int ib_register_device(struct ib_device *device, const char *name,
2646 int (*port_callback)(struct ib_device *, u8,
2648 void ib_unregister_device(struct ib_device *device);
2650 int ib_register_client (struct ib_client *client);
2651 void ib_unregister_client(struct ib_client *client);
2653 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2654 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2657 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2658 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2659 unsigned long pfn, unsigned long size, pgprot_t prot);
2660 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2661 struct vm_area_struct *vma, struct page *page,
2662 unsigned long size);
2664 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2665 struct vm_area_struct *vma,
2666 unsigned long pfn, unsigned long size,
2671 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2672 struct vm_area_struct *vma, struct page *page,
2679 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2681 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2684 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2686 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2689 static inline bool ib_is_buffer_cleared(const void __user *p,
2695 if (len > USHRT_MAX)
2698 buf = memdup_user(p, len);
2702 ret = !memchr_inv(buf, 0, len);
2707 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2711 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2715 * ib_is_destroy_retryable - Check whether the uobject destruction
2717 * @ret: The initial destruction return code
2718 * @why: remove reason
2719 * @uobj: The uobject that is destroyed
2721 * This function is a helper function that IB layer and low-level drivers
2722 * can use to consider whether the destruction of the given uobject is
2724 * It checks the original return code, if it wasn't success the destruction
2725 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2726 * the remove reason. (i.e. why).
2727 * Must be called with the object locked for destroy.
2729 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2730 struct ib_uobject *uobj)
2732 return ret && (why == RDMA_REMOVE_DESTROY ||
2733 uobj->context->cleanup_retryable);
2737 * ib_destroy_usecnt - Called during destruction to check the usecnt
2738 * @usecnt: The usecnt atomic
2739 * @why: remove reason
2740 * @uobj: The uobject that is destroyed
2742 * Non-zero usecnts will block destruction unless destruction was triggered by
2743 * a ucontext cleanup.
2745 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2746 enum rdma_remove_reason why,
2747 struct ib_uobject *uobj)
2749 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2755 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2756 * contains all required attributes and no attributes not allowed for
2757 * the given QP state transition.
2758 * @cur_state: Current QP state
2759 * @next_state: Next QP state
2761 * @mask: Mask of supplied QP attributes
2763 * This function is a helper function that a low-level driver's
2764 * modify_qp method can use to validate the consumer's input. It
2765 * checks that cur_state and next_state are valid QP states, that a
2766 * transition from cur_state to next_state is allowed by the IB spec,
2767 * and that the attribute mask supplied is allowed for the transition.
2769 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2770 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2772 void ib_register_event_handler(struct ib_event_handler *event_handler);
2773 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2774 void ib_dispatch_event(struct ib_event *event);
2776 int ib_query_port(struct ib_device *device,
2777 u8 port_num, struct ib_port_attr *port_attr);
2779 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2783 * rdma_cap_ib_switch - Check if the device is IB switch
2784 * @device: Device to check
2786 * Device driver is responsible for setting is_switch bit on
2787 * in ib_device structure at init time.
2789 * Return: true if the device is IB switch.
2791 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2793 return device->is_switch;
2797 * rdma_start_port - Return the first valid port number for the device
2800 * @device: Device to be checked
2802 * Return start port number
2804 static inline u8 rdma_start_port(const struct ib_device *device)
2806 return rdma_cap_ib_switch(device) ? 0 : 1;
2810 * rdma_end_port - Return the last valid port number for the device
2813 * @device: Device to be checked
2815 * Return last port number
2817 static inline u8 rdma_end_port(const struct ib_device *device)
2819 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2822 static inline int rdma_is_port_valid(const struct ib_device *device,
2825 return (port >= rdma_start_port(device) &&
2826 port <= rdma_end_port(device));
2829 static inline bool rdma_is_grh_required(const struct ib_device *device,
2832 return device->port_immutable[port_num].core_cap_flags &
2833 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2836 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2838 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2841 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2843 return device->port_immutable[port_num].core_cap_flags &
2844 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2847 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2849 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2852 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2854 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2857 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2859 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2862 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2864 return rdma_protocol_ib(device, port_num) ||
2865 rdma_protocol_roce(device, port_num);
2868 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2870 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2873 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2875 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2879 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2880 * Management Datagrams.
2881 * @device: Device to check
2882 * @port_num: Port number to check
2884 * Management Datagrams (MAD) are a required part of the InfiniBand
2885 * specification and are supported on all InfiniBand devices. A slightly
2886 * extended version are also supported on OPA interfaces.
2888 * Return: true if the port supports sending/receiving of MAD packets.
2890 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2892 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2896 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2897 * Management Datagrams.
2898 * @device: Device to check
2899 * @port_num: Port number to check
2901 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2902 * datagrams with their own versions. These OPA MADs share many but not all of
2903 * the characteristics of InfiniBand MADs.
2905 * OPA MADs differ in the following ways:
2907 * 1) MADs are variable size up to 2K
2908 * IBTA defined MADs remain fixed at 256 bytes
2909 * 2) OPA SMPs must carry valid PKeys
2910 * 3) OPA SMP packets are a different format
2912 * Return: true if the port supports OPA MAD packet formats.
2914 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2916 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2917 == RDMA_CORE_CAP_OPA_MAD;
2921 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2922 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2923 * @device: Device to check
2924 * @port_num: Port number to check
2926 * Each InfiniBand node is required to provide a Subnet Management Agent
2927 * that the subnet manager can access. Prior to the fabric being fully
2928 * configured by the subnet manager, the SMA is accessed via a well known
2929 * interface called the Subnet Management Interface (SMI). This interface
2930 * uses directed route packets to communicate with the SM to get around the
2931 * chicken and egg problem of the SM needing to know what's on the fabric
2932 * in order to configure the fabric, and needing to configure the fabric in
2933 * order to send packets to the devices on the fabric. These directed
2934 * route packets do not need the fabric fully configured in order to reach
2935 * their destination. The SMI is the only method allowed to send
2936 * directed route packets on an InfiniBand fabric.
2938 * Return: true if the port provides an SMI.
2940 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2942 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2946 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2947 * Communication Manager.
2948 * @device: Device to check
2949 * @port_num: Port number to check
2951 * The InfiniBand Communication Manager is one of many pre-defined General
2952 * Service Agents (GSA) that are accessed via the General Service
2953 * Interface (GSI). It's role is to facilitate establishment of connections
2954 * between nodes as well as other management related tasks for established
2957 * Return: true if the port supports an IB CM (this does not guarantee that
2958 * a CM is actually running however).
2960 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2962 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2966 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2967 * Communication Manager.
2968 * @device: Device to check
2969 * @port_num: Port number to check
2971 * Similar to above, but specific to iWARP connections which have a different
2972 * managment protocol than InfiniBand.
2974 * Return: true if the port supports an iWARP CM (this does not guarantee that
2975 * a CM is actually running however).
2977 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2979 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2983 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2984 * Subnet Administration.
2985 * @device: Device to check
2986 * @port_num: Port number to check
2988 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2989 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2990 * fabrics, devices should resolve routes to other hosts by contacting the
2991 * SA to query the proper route.
2993 * Return: true if the port should act as a client to the fabric Subnet
2994 * Administration interface. This does not imply that the SA service is
2997 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2999 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
3003 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3005 * @device: Device to check
3006 * @port_num: Port number to check
3008 * InfiniBand multicast registration is more complex than normal IPv4 or
3009 * IPv6 multicast registration. Each Host Channel Adapter must register
3010 * with the Subnet Manager when it wishes to join a multicast group. It
3011 * should do so only once regardless of how many queue pairs it subscribes
3012 * to this group. And it should leave the group only after all queue pairs
3013 * attached to the group have been detached.
3015 * Return: true if the port must undertake the additional adminstrative
3016 * overhead of registering/unregistering with the SM and tracking of the
3017 * total number of queue pairs attached to the multicast group.
3019 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3021 return rdma_cap_ib_sa(device, port_num);
3025 * rdma_cap_af_ib - Check if the port of device has the capability
3026 * Native Infiniband Address.
3027 * @device: Device to check
3028 * @port_num: Port number to check
3030 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3031 * GID. RoCE uses a different mechanism, but still generates a GID via
3032 * a prescribed mechanism and port specific data.
3034 * Return: true if the port uses a GID address to identify devices on the
3037 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3039 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3043 * rdma_cap_eth_ah - Check if the port of device has the capability
3044 * Ethernet Address Handle.
3045 * @device: Device to check
3046 * @port_num: Port number to check
3048 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3049 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3050 * port. Normally, packet headers are generated by the sending host
3051 * adapter, but when sending connectionless datagrams, we must manually
3052 * inject the proper headers for the fabric we are communicating over.
3054 * Return: true if we are running as a RoCE port and must force the
3055 * addition of a Global Route Header built from our Ethernet Address
3056 * Handle into our header list for connectionless packets.
3058 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3060 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3064 * rdma_cap_opa_ah - Check if the port of device supports
3065 * OPA Address handles
3066 * @device: Device to check
3067 * @port_num: Port number to check
3069 * Return: true if we are running on an OPA device which supports
3070 * the extended OPA addressing.
3072 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3074 return (device->port_immutable[port_num].core_cap_flags &
3075 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3079 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3082 * @port_num: Port number
3084 * This MAD size includes the MAD headers and MAD payload. No other headers
3087 * Return the max MAD size required by the Port. Will return 0 if the port
3088 * does not support MADs
3090 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3092 return device->port_immutable[port_num].max_mad_size;
3096 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3097 * @device: Device to check
3098 * @port_num: Port number to check
3100 * RoCE GID table mechanism manages the various GIDs for a device.
3102 * NOTE: if allocating the port's GID table has failed, this call will still
3103 * return true, but any RoCE GID table API will fail.
3105 * Return: true if the port uses RoCE GID table mechanism in order to manage
3108 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3111 return rdma_protocol_roce(device, port_num) &&
3112 device->add_gid && device->del_gid;
3116 * Check if the device supports READ W/ INVALIDATE.
3118 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3121 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3122 * has support for it yet.
3124 return rdma_protocol_iwarp(dev, port_num);
3127 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3129 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3130 struct ifla_vf_info *info);
3131 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3132 struct ifla_vf_stats *stats);
3133 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3136 int ib_query_pkey(struct ib_device *device,
3137 u8 port_num, u16 index, u16 *pkey);
3139 int ib_modify_device(struct ib_device *device,
3140 int device_modify_mask,
3141 struct ib_device_modify *device_modify);
3143 int ib_modify_port(struct ib_device *device,
3144 u8 port_num, int port_modify_mask,
3145 struct ib_port_modify *port_modify);
3147 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3148 u8 *port_num, u16 *index);
3150 int ib_find_pkey(struct ib_device *device,
3151 u8 port_num, u16 pkey, u16 *index);
3155 * Create a memory registration for all memory in the system and place
3156 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3157 * ULPs to avoid the overhead of dynamic MRs.
3159 * This flag is generally considered unsafe and must only be used in
3160 * extremly trusted environments. Every use of it will log a warning
3161 * in the kernel log.
3163 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3166 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3167 const char *caller);
3168 #define ib_alloc_pd(device, flags) \
3169 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3170 void ib_dealloc_pd(struct ib_pd *pd);
3173 * rdma_create_ah - Creates an address handle for the given address vector.
3174 * @pd: The protection domain associated with the address handle.
3175 * @ah_attr: The attributes of the address vector.
3177 * The address handle is used to reference a local or global destination
3178 * in all UD QP post sends.
3180 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3183 * rdma_create_user_ah - Creates an address handle for the given address vector.
3184 * It resolves destination mac address for ah attribute of RoCE type.
3185 * @pd: The protection domain associated with the address handle.
3186 * @ah_attr: The attributes of the address vector.
3187 * @udata: pointer to user's input output buffer information need by
3190 * It returns 0 on success and returns appropriate error code on error.
3191 * The address handle is used to reference a local or global destination
3192 * in all UD QP post sends.
3194 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3195 struct rdma_ah_attr *ah_attr,
3196 struct ib_udata *udata);
3198 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3200 * @hdr: the L3 header to parse
3201 * @net_type: type of header to parse
3202 * @sgid: place to store source gid
3203 * @dgid: place to store destination gid
3205 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3206 enum rdma_network_type net_type,
3207 union ib_gid *sgid, union ib_gid *dgid);
3210 * ib_get_rdma_header_version - Get the header version
3211 * @hdr: the L3 header to parse
3213 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3216 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3218 * @device: Device on which the received message arrived.
3219 * @port_num: Port on which the received message arrived.
3220 * @wc: Work completion associated with the received message.
3221 * @grh: References the received global route header. This parameter is
3222 * ignored unless the work completion indicates that the GRH is valid.
3223 * @ah_attr: Returned attributes that can be used when creating an address
3224 * handle for replying to the message.
3225 * When ib_init_ah_attr_from_wc() returns success,
3226 * (a) for IB link layer it optionally contains a reference to SGID attribute
3227 * when GRH is present for IB link layer.
3228 * (b) for RoCE link layer it contains a reference to SGID attribute.
3229 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3230 * attributes which are initialized using ib_init_ah_attr_from_wc().
3233 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3234 const struct ib_wc *wc, const struct ib_grh *grh,
3235 struct rdma_ah_attr *ah_attr);
3238 * ib_create_ah_from_wc - Creates an address handle associated with the
3239 * sender of the specified work completion.
3240 * @pd: The protection domain associated with the address handle.
3241 * @wc: Work completion information associated with a received message.
3242 * @grh: References the received global route header. This parameter is
3243 * ignored unless the work completion indicates that the GRH is valid.
3244 * @port_num: The outbound port number to associate with the address.
3246 * The address handle is used to reference a local or global destination
3247 * in all UD QP post sends.
3249 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3250 const struct ib_grh *grh, u8 port_num);
3253 * rdma_modify_ah - Modifies the address vector associated with an address
3255 * @ah: The address handle to modify.
3256 * @ah_attr: The new address vector attributes to associate with the
3259 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3262 * rdma_query_ah - Queries the address vector associated with an address
3264 * @ah: The address handle to query.
3265 * @ah_attr: The address vector attributes associated with the address
3268 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3271 * rdma_destroy_ah - Destroys an address handle.
3272 * @ah: The address handle to destroy.
3274 int rdma_destroy_ah(struct ib_ah *ah);
3277 * ib_create_srq - Creates a SRQ associated with the specified protection
3279 * @pd: The protection domain associated with the SRQ.
3280 * @srq_init_attr: A list of initial attributes required to create the
3281 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3282 * the actual capabilities of the created SRQ.
3284 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3285 * requested size of the SRQ, and set to the actual values allocated
3286 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3287 * will always be at least as large as the requested values.
3289 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3290 struct ib_srq_init_attr *srq_init_attr);
3293 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3294 * @srq: The SRQ to modify.
3295 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3296 * the current values of selected SRQ attributes are returned.
3297 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3298 * are being modified.
3300 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3301 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3302 * the number of receives queued drops below the limit.
3304 int ib_modify_srq(struct ib_srq *srq,
3305 struct ib_srq_attr *srq_attr,
3306 enum ib_srq_attr_mask srq_attr_mask);
3309 * ib_query_srq - Returns the attribute list and current values for the
3311 * @srq: The SRQ to query.
3312 * @srq_attr: The attributes of the specified SRQ.
3314 int ib_query_srq(struct ib_srq *srq,
3315 struct ib_srq_attr *srq_attr);
3318 * ib_destroy_srq - Destroys the specified SRQ.
3319 * @srq: The SRQ to destroy.
3321 int ib_destroy_srq(struct ib_srq *srq);
3324 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3325 * @srq: The SRQ to post the work request on.
3326 * @recv_wr: A list of work requests to post on the receive queue.
3327 * @bad_recv_wr: On an immediate failure, this parameter will reference
3328 * the work request that failed to be posted on the QP.
3330 static inline int ib_post_srq_recv(struct ib_srq *srq,
3331 const struct ib_recv_wr *recv_wr,
3332 const struct ib_recv_wr **bad_recv_wr)
3334 const struct ib_recv_wr *dummy;
3336 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr ? : &dummy);
3340 * ib_create_qp - Creates a QP associated with the specified protection
3342 * @pd: The protection domain associated with the QP.
3343 * @qp_init_attr: A list of initial attributes required to create the
3344 * QP. If QP creation succeeds, then the attributes are updated to
3345 * the actual capabilities of the created QP.
3347 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3348 struct ib_qp_init_attr *qp_init_attr);
3351 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3352 * @qp: The QP to modify.
3353 * @attr: On input, specifies the QP attributes to modify. On output,
3354 * the current values of selected QP attributes are returned.
3355 * @attr_mask: A bit-mask used to specify which attributes of the QP
3356 * are being modified.
3357 * @udata: pointer to user's input output buffer information
3358 * are being modified.
3359 * It returns 0 on success and returns appropriate error code on error.
3361 int ib_modify_qp_with_udata(struct ib_qp *qp,
3362 struct ib_qp_attr *attr,
3364 struct ib_udata *udata);
3367 * ib_modify_qp - Modifies the attributes for the specified QP and then
3368 * transitions the QP to the given state.
3369 * @qp: The QP to modify.
3370 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3371 * the current values of selected QP attributes are returned.
3372 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3373 * are being modified.
3375 int ib_modify_qp(struct ib_qp *qp,
3376 struct ib_qp_attr *qp_attr,
3380 * ib_query_qp - Returns the attribute list and current values for the
3382 * @qp: The QP to query.
3383 * @qp_attr: The attributes of the specified QP.
3384 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3385 * @qp_init_attr: Additional attributes of the selected QP.
3387 * The qp_attr_mask may be used to limit the query to gathering only the
3388 * selected attributes.
3390 int ib_query_qp(struct ib_qp *qp,
3391 struct ib_qp_attr *qp_attr,
3393 struct ib_qp_init_attr *qp_init_attr);
3396 * ib_destroy_qp - Destroys the specified QP.
3397 * @qp: The QP to destroy.
3399 int ib_destroy_qp(struct ib_qp *qp);
3402 * ib_open_qp - Obtain a reference to an existing sharable QP.
3403 * @xrcd - XRC domain
3404 * @qp_open_attr: Attributes identifying the QP to open.
3406 * Returns a reference to a sharable QP.
3408 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3409 struct ib_qp_open_attr *qp_open_attr);
3412 * ib_close_qp - Release an external reference to a QP.
3413 * @qp: The QP handle to release
3415 * The opened QP handle is released by the caller. The underlying
3416 * shared QP is not destroyed until all internal references are released.
3418 int ib_close_qp(struct ib_qp *qp);
3421 * ib_post_send - Posts a list of work requests to the send queue of
3423 * @qp: The QP to post the work request on.
3424 * @send_wr: A list of work requests to post on the send queue.
3425 * @bad_send_wr: On an immediate failure, this parameter will reference
3426 * the work request that failed to be posted on the QP.
3428 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3429 * error is returned, the QP state shall not be affected,
3430 * ib_post_send() will return an immediate error after queueing any
3431 * earlier work requests in the list.
3433 static inline int ib_post_send(struct ib_qp *qp,
3434 const struct ib_send_wr *send_wr,
3435 const struct ib_send_wr **bad_send_wr)
3437 const struct ib_send_wr *dummy;
3439 return qp->device->post_send(qp, send_wr, bad_send_wr ? : &dummy);
3443 * ib_post_recv - Posts a list of work requests to the receive queue of
3445 * @qp: The QP to post the work request on.
3446 * @recv_wr: A list of work requests to post on the receive queue.
3447 * @bad_recv_wr: On an immediate failure, this parameter will reference
3448 * the work request that failed to be posted on the QP.
3450 static inline int ib_post_recv(struct ib_qp *qp,
3451 const struct ib_recv_wr *recv_wr,
3452 const struct ib_recv_wr **bad_recv_wr)
3454 const struct ib_recv_wr *dummy;
3456 return qp->device->post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3459 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3460 int nr_cqe, int comp_vector,
3461 enum ib_poll_context poll_ctx, const char *caller);
3462 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3463 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3465 void ib_free_cq(struct ib_cq *cq);
3466 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3469 * ib_create_cq - Creates a CQ on the specified device.
3470 * @device: The device on which to create the CQ.
3471 * @comp_handler: A user-specified callback that is invoked when a
3472 * completion event occurs on the CQ.
3473 * @event_handler: A user-specified callback that is invoked when an
3474 * asynchronous event not associated with a completion occurs on the CQ.
3475 * @cq_context: Context associated with the CQ returned to the user via
3476 * the associated completion and event handlers.
3477 * @cq_attr: The attributes the CQ should be created upon.
3479 * Users can examine the cq structure to determine the actual CQ size.
3481 struct ib_cq *__ib_create_cq(struct ib_device *device,
3482 ib_comp_handler comp_handler,
3483 void (*event_handler)(struct ib_event *, void *),
3485 const struct ib_cq_init_attr *cq_attr,
3486 const char *caller);
3487 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3488 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3491 * ib_resize_cq - Modifies the capacity of the CQ.
3492 * @cq: The CQ to resize.
3493 * @cqe: The minimum size of the CQ.
3495 * Users can examine the cq structure to determine the actual CQ size.
3497 int ib_resize_cq(struct ib_cq *cq, int cqe);
3500 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3501 * @cq: The CQ to modify.
3502 * @cq_count: number of CQEs that will trigger an event
3503 * @cq_period: max period of time in usec before triggering an event
3506 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3509 * ib_destroy_cq - Destroys the specified CQ.
3510 * @cq: The CQ to destroy.
3512 int ib_destroy_cq(struct ib_cq *cq);
3515 * ib_poll_cq - poll a CQ for completion(s)
3516 * @cq:the CQ being polled
3517 * @num_entries:maximum number of completions to return
3518 * @wc:array of at least @num_entries &struct ib_wc where completions
3521 * Poll a CQ for (possibly multiple) completions. If the return value
3522 * is < 0, an error occurred. If the return value is >= 0, it is the
3523 * number of completions returned. If the return value is
3524 * non-negative and < num_entries, then the CQ was emptied.
3526 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3529 return cq->device->poll_cq(cq, num_entries, wc);
3533 * ib_req_notify_cq - Request completion notification on a CQ.
3534 * @cq: The CQ to generate an event for.
3536 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3537 * to request an event on the next solicited event or next work
3538 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3539 * may also be |ed in to request a hint about missed events, as
3543 * < 0 means an error occurred while requesting notification
3544 * == 0 means notification was requested successfully, and if
3545 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3546 * were missed and it is safe to wait for another event. In
3547 * this case is it guaranteed that any work completions added
3548 * to the CQ since the last CQ poll will trigger a completion
3549 * notification event.
3550 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3551 * in. It means that the consumer must poll the CQ again to
3552 * make sure it is empty to avoid missing an event because of a
3553 * race between requesting notification and an entry being
3554 * added to the CQ. This return value means it is possible
3555 * (but not guaranteed) that a work completion has been added
3556 * to the CQ since the last poll without triggering a
3557 * completion notification event.
3559 static inline int ib_req_notify_cq(struct ib_cq *cq,
3560 enum ib_cq_notify_flags flags)
3562 return cq->device->req_notify_cq(cq, flags);
3566 * ib_req_ncomp_notif - Request completion notification when there are
3567 * at least the specified number of unreaped completions on the CQ.
3568 * @cq: The CQ to generate an event for.
3569 * @wc_cnt: The number of unreaped completions that should be on the
3570 * CQ before an event is generated.
3572 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3574 return cq->device->req_ncomp_notif ?
3575 cq->device->req_ncomp_notif(cq, wc_cnt) :
3580 * ib_dma_mapping_error - check a DMA addr for error
3581 * @dev: The device for which the dma_addr was created
3582 * @dma_addr: The DMA address to check
3584 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3586 return dma_mapping_error(dev->dma_device, dma_addr);
3590 * ib_dma_map_single - Map a kernel virtual address to DMA address
3591 * @dev: The device for which the dma_addr is to be created
3592 * @cpu_addr: The kernel virtual address
3593 * @size: The size of the region in bytes
3594 * @direction: The direction of the DMA
3596 static inline u64 ib_dma_map_single(struct ib_device *dev,
3597 void *cpu_addr, size_t size,
3598 enum dma_data_direction direction)
3600 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3604 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3605 * @dev: The device for which the DMA address was created
3606 * @addr: The DMA address
3607 * @size: The size of the region in bytes
3608 * @direction: The direction of the DMA
3610 static inline void ib_dma_unmap_single(struct ib_device *dev,
3611 u64 addr, size_t size,
3612 enum dma_data_direction direction)
3614 dma_unmap_single(dev->dma_device, addr, size, direction);
3618 * ib_dma_map_page - Map a physical page to DMA address
3619 * @dev: The device for which the dma_addr is to be created
3620 * @page: The page to be mapped
3621 * @offset: The offset within the page
3622 * @size: The size of the region in bytes
3623 * @direction: The direction of the DMA
3625 static inline u64 ib_dma_map_page(struct ib_device *dev,
3627 unsigned long offset,
3629 enum dma_data_direction direction)
3631 return dma_map_page(dev->dma_device, page, offset, size, direction);
3635 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3636 * @dev: The device for which the DMA address was created
3637 * @addr: The DMA address
3638 * @size: The size of the region in bytes
3639 * @direction: The direction of the DMA
3641 static inline void ib_dma_unmap_page(struct ib_device *dev,
3642 u64 addr, size_t size,
3643 enum dma_data_direction direction)
3645 dma_unmap_page(dev->dma_device, addr, size, direction);
3649 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3650 * @dev: The device for which the DMA addresses are to be created
3651 * @sg: The array of scatter/gather entries
3652 * @nents: The number of scatter/gather entries
3653 * @direction: The direction of the DMA
3655 static inline int ib_dma_map_sg(struct ib_device *dev,
3656 struct scatterlist *sg, int nents,
3657 enum dma_data_direction direction)
3659 return dma_map_sg(dev->dma_device, sg, nents, direction);
3663 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3664 * @dev: The device for which the DMA addresses were created
3665 * @sg: The array of scatter/gather entries
3666 * @nents: The number of scatter/gather entries
3667 * @direction: The direction of the DMA
3669 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3670 struct scatterlist *sg, int nents,
3671 enum dma_data_direction direction)
3673 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3676 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3677 struct scatterlist *sg, int nents,
3678 enum dma_data_direction direction,
3679 unsigned long dma_attrs)
3681 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3685 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3686 struct scatterlist *sg, int nents,
3687 enum dma_data_direction direction,
3688 unsigned long dma_attrs)
3690 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3693 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3694 * @dev: The device for which the DMA addresses were created
3695 * @sg: The scatter/gather entry
3697 * Note: this function is obsolete. To do: change all occurrences of
3698 * ib_sg_dma_address() into sg_dma_address().
3700 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3701 struct scatterlist *sg)
3703 return sg_dma_address(sg);
3707 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3708 * @dev: The device for which the DMA addresses were created
3709 * @sg: The scatter/gather entry
3711 * Note: this function is obsolete. To do: change all occurrences of
3712 * ib_sg_dma_len() into sg_dma_len().
3714 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3715 struct scatterlist *sg)
3717 return sg_dma_len(sg);
3721 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3722 * @dev: The device for which the DMA address was created
3723 * @addr: The DMA address
3724 * @size: The size of the region in bytes
3725 * @dir: The direction of the DMA
3727 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3730 enum dma_data_direction dir)
3732 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3736 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3737 * @dev: The device for which the DMA address was created
3738 * @addr: The DMA address
3739 * @size: The size of the region in bytes
3740 * @dir: The direction of the DMA
3742 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3745 enum dma_data_direction dir)
3747 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3751 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3752 * @dev: The device for which the DMA address is requested
3753 * @size: The size of the region to allocate in bytes
3754 * @dma_handle: A pointer for returning the DMA address of the region
3755 * @flag: memory allocator flags
3757 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3759 dma_addr_t *dma_handle,
3762 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3766 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3767 * @dev: The device for which the DMA addresses were allocated
3768 * @size: The size of the region
3769 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3770 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3772 static inline void ib_dma_free_coherent(struct ib_device *dev,
3773 size_t size, void *cpu_addr,
3774 dma_addr_t dma_handle)
3776 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3780 * ib_dereg_mr - Deregisters a memory region and removes it from the
3781 * HCA translation table.
3782 * @mr: The memory region to deregister.
3784 * This function can fail, if the memory region has memory windows bound to it.
3786 int ib_dereg_mr(struct ib_mr *mr);
3788 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3789 enum ib_mr_type mr_type,
3793 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3795 * @mr - struct ib_mr pointer to be updated.
3796 * @newkey - new key to be used.
3798 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3800 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3801 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3805 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3806 * for calculating a new rkey for type 2 memory windows.
3807 * @rkey - the rkey to increment.
3809 static inline u32 ib_inc_rkey(u32 rkey)
3811 const u32 mask = 0x000000ff;
3812 return ((rkey + 1) & mask) | (rkey & ~mask);
3816 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3817 * @pd: The protection domain associated with the unmapped region.
3818 * @mr_access_flags: Specifies the memory access rights.
3819 * @fmr_attr: Attributes of the unmapped region.
3821 * A fast memory region must be mapped before it can be used as part of
3824 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3825 int mr_access_flags,
3826 struct ib_fmr_attr *fmr_attr);
3829 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3830 * @fmr: The fast memory region to associate with the pages.
3831 * @page_list: An array of physical pages to map to the fast memory region.
3832 * @list_len: The number of pages in page_list.
3833 * @iova: The I/O virtual address to use with the mapped region.
3835 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3836 u64 *page_list, int list_len,
3839 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3843 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3844 * @fmr_list: A linked list of fast memory regions to unmap.
3846 int ib_unmap_fmr(struct list_head *fmr_list);
3849 * ib_dealloc_fmr - Deallocates a fast memory region.
3850 * @fmr: The fast memory region to deallocate.
3852 int ib_dealloc_fmr(struct ib_fmr *fmr);
3855 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3856 * @qp: QP to attach to the multicast group. The QP must be type
3858 * @gid: Multicast group GID.
3859 * @lid: Multicast group LID in host byte order.
3861 * In order to send and receive multicast packets, subnet
3862 * administration must have created the multicast group and configured
3863 * the fabric appropriately. The port associated with the specified
3864 * QP must also be a member of the multicast group.
3866 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3869 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3870 * @qp: QP to detach from the multicast group.
3871 * @gid: Multicast group GID.
3872 * @lid: Multicast group LID in host byte order.
3874 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3877 * ib_alloc_xrcd - Allocates an XRC domain.
3878 * @device: The device on which to allocate the XRC domain.
3879 * @caller: Module name for kernel consumers
3881 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3882 #define ib_alloc_xrcd(device) \
3883 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3886 * ib_dealloc_xrcd - Deallocates an XRC domain.
3887 * @xrcd: The XRC domain to deallocate.
3889 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3891 static inline int ib_check_mr_access(int flags)
3894 * Local write permission is required if remote write or
3895 * remote atomic permission is also requested.
3897 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3898 !(flags & IB_ACCESS_LOCAL_WRITE))
3904 static inline bool ib_access_writable(int access_flags)
3907 * We have writable memory backing the MR if any of the following
3908 * access flags are set. "Local write" and "remote write" obviously
3909 * require write access. "Remote atomic" can do things like fetch and
3910 * add, which will modify memory, and "MW bind" can change permissions
3911 * by binding a window.
3913 return access_flags &
3914 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3915 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3919 * ib_check_mr_status: lightweight check of MR status.
3920 * This routine may provide status checks on a selected
3921 * ib_mr. first use is for signature status check.
3923 * @mr: A memory region.
3924 * @check_mask: Bitmask of which checks to perform from
3925 * ib_mr_status_check enumeration.
3926 * @mr_status: The container of relevant status checks.
3927 * failed checks will be indicated in the status bitmask
3928 * and the relevant info shall be in the error item.
3930 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3931 struct ib_mr_status *mr_status);
3933 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3934 u16 pkey, const union ib_gid *gid,
3935 const struct sockaddr *addr);
3936 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3937 struct ib_wq_init_attr *init_attr);
3938 int ib_destroy_wq(struct ib_wq *wq);
3939 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3941 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3942 struct ib_rwq_ind_table_init_attr*
3943 wq_ind_table_init_attr);
3944 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3946 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3947 unsigned int *sg_offset, unsigned int page_size);
3950 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3951 unsigned int *sg_offset, unsigned int page_size)
3955 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3961 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3962 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3964 void ib_drain_rq(struct ib_qp *qp);
3965 void ib_drain_sq(struct ib_qp *qp);
3966 void ib_drain_qp(struct ib_qp *qp);
3968 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3970 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3972 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3973 return attr->roce.dmac;
3977 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3979 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3980 attr->ib.dlid = (u16)dlid;
3981 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3982 attr->opa.dlid = dlid;
3985 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3987 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3988 return attr->ib.dlid;
3989 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3990 return attr->opa.dlid;
3994 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3999 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4004 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4007 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4008 attr->ib.src_path_bits = src_path_bits;
4009 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4010 attr->opa.src_path_bits = src_path_bits;
4013 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4015 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4016 return attr->ib.src_path_bits;
4017 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4018 return attr->opa.src_path_bits;
4022 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4025 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4026 attr->opa.make_grd = make_grd;
4029 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4031 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4032 return attr->opa.make_grd;
4036 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4038 attr->port_num = port_num;
4041 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4043 return attr->port_num;
4046 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4049 attr->static_rate = static_rate;
4052 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4054 return attr->static_rate;
4057 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4058 enum ib_ah_flags flag)
4060 attr->ah_flags = flag;
4063 static inline enum ib_ah_flags
4064 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4066 return attr->ah_flags;
4069 static inline const struct ib_global_route
4070 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4075 /*To retrieve and modify the grh */
4076 static inline struct ib_global_route
4077 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4082 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4084 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4086 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4089 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4092 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4094 grh->dgid.global.subnet_prefix = prefix;
4097 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4100 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4102 grh->dgid.global.interface_id = if_id;
4105 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4106 union ib_gid *dgid, u32 flow_label,
4107 u8 sgid_index, u8 hop_limit,
4110 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4112 attr->ah_flags = IB_AH_GRH;
4115 grh->flow_label = flow_label;
4116 grh->sgid_index = sgid_index;
4117 grh->hop_limit = hop_limit;
4118 grh->traffic_class = traffic_class;
4119 grh->sgid_attr = NULL;
4122 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4123 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4124 u32 flow_label, u8 hop_limit, u8 traffic_class,
4125 const struct ib_gid_attr *sgid_attr);
4126 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4127 const struct rdma_ah_attr *src);
4128 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4129 const struct rdma_ah_attr *new);
4130 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4133 * rdma_ah_find_type - Return address handle type.
4135 * @dev: Device to be checked
4136 * @port_num: Port number
4138 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4141 if (rdma_protocol_roce(dev, port_num))
4142 return RDMA_AH_ATTR_TYPE_ROCE;
4143 if (rdma_protocol_ib(dev, port_num)) {
4144 if (rdma_cap_opa_ah(dev, port_num))
4145 return RDMA_AH_ATTR_TYPE_OPA;
4146 return RDMA_AH_ATTR_TYPE_IB;
4149 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4153 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4154 * In the current implementation the only way to get
4155 * get the 32bit lid is from other sources for OPA.
4156 * For IB, lids will always be 16bits so cast the
4157 * value accordingly.
4161 static inline u16 ib_lid_cpu16(u32 lid)
4163 WARN_ON_ONCE(lid & 0xFFFF0000);
4168 * ib_lid_be16 - Return lid in 16bit BE encoding.
4172 static inline __be16 ib_lid_be16(u32 lid)
4174 WARN_ON_ONCE(lid & 0xFFFF0000);
4175 return cpu_to_be16((u16)lid);
4179 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4181 * @device: the rdma device
4182 * @comp_vector: index of completion vector
4184 * Returns NULL on failure, otherwise a corresponding cpu map of the
4185 * completion vector (returns all-cpus map if the device driver doesn't
4186 * implement get_vector_affinity).
4188 static inline const struct cpumask *
4189 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4191 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4192 !device->get_vector_affinity)
4195 return device->get_vector_affinity(device, comp_vector);
4200 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4201 * and add their gids, as needed, to the relevant RoCE devices.
4203 * @device: the rdma device
4205 void rdma_roce_rescan_device(struct ib_device *ibdev);
4207 struct ib_ucontext *ib_uverbs_get_ucontext(struct ib_uverbs_file *ufile);
4209 int uverbs_destroy_def_handler(struct ib_uverbs_file *file,
4210 struct uverbs_attr_bundle *attrs);
4212 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4213 enum rdma_netdev_t type, const char *name,
4214 unsigned char name_assign_type,
4215 void (*setup)(struct net_device *));
4217 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4218 enum rdma_netdev_t type, const char *name,
4219 unsigned char name_assign_type,
4220 void (*setup)(struct net_device *),
4221 struct net_device *netdev);
4224 * rdma_set_device_sysfs_group - Set device attributes group to have
4225 * driver specific sysfs entries at
4226 * for infiniband class.
4228 * @device: device pointer for which attributes to be created
4229 * @group: Pointer to group which should be added when device
4230 * is registered with sysfs.
4231 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4232 * group per device to have sysfs attributes.
4234 * NOTE: New drivers should not make use of this API; instead new device
4235 * parameter should be exposed via netlink command. This API and mechanism
4236 * exist only for existing drivers.
4239 rdma_set_device_sysfs_group(struct ib_device *dev,
4240 const struct attribute_group *group)
4242 dev->groups[1] = group;
4245 #endif /* IB_VERBS_H */