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>
66 extern struct workqueue_struct *ib_wq;
67 extern struct workqueue_struct *ib_comp_wq;
77 extern union ib_gid zgid;
80 /* If link layer is Ethernet, this is RoCE V1 */
83 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
87 #define ROCE_V2_UDP_DPORT 4791
89 enum ib_gid_type gid_type;
90 struct net_device *ndev;
94 /* IB values map to NodeInfo:NodeType. */
104 /* set the local administered indication */
105 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
108 enum rdma_transport_type {
110 RDMA_TRANSPORT_IWARP,
111 RDMA_TRANSPORT_USNIC,
112 RDMA_TRANSPORT_USNIC_UDP
115 enum rdma_protocol_type {
119 RDMA_PROTOCOL_USNIC_UDP
122 __attribute_const__ enum rdma_transport_type
123 rdma_node_get_transport(enum rdma_node_type node_type);
125 enum rdma_network_type {
127 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
132 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
134 if (network_type == RDMA_NETWORK_IPV4 ||
135 network_type == RDMA_NETWORK_IPV6)
136 return IB_GID_TYPE_ROCE_UDP_ENCAP;
138 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
139 return IB_GID_TYPE_IB;
142 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
145 if (gid_type == IB_GID_TYPE_IB)
146 return RDMA_NETWORK_IB;
148 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
149 return RDMA_NETWORK_IPV4;
151 return RDMA_NETWORK_IPV6;
154 enum rdma_link_layer {
155 IB_LINK_LAYER_UNSPECIFIED,
156 IB_LINK_LAYER_INFINIBAND,
157 IB_LINK_LAYER_ETHERNET,
160 enum ib_device_cap_flags {
161 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
162 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
163 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
164 IB_DEVICE_RAW_MULTI = (1 << 3),
165 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
166 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
167 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
168 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
169 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
170 IB_DEVICE_INIT_TYPE = (1 << 9),
171 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
172 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
173 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
174 IB_DEVICE_SRQ_RESIZE = (1 << 13),
175 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
178 * This device supports a per-device lkey or stag that can be
179 * used without performing a memory registration for the local
180 * memory. Note that ULPs should never check this flag, but
181 * instead of use the local_dma_lkey flag in the ib_pd structure,
182 * which will always contain a usable lkey.
184 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
185 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16),
186 IB_DEVICE_MEM_WINDOW = (1 << 17),
188 * Devices should set IB_DEVICE_UD_IP_SUM if they support
189 * insertion of UDP and TCP checksum on outgoing UD IPoIB
190 * messages and can verify the validity of checksum for
191 * incoming messages. Setting this flag implies that the
192 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
194 IB_DEVICE_UD_IP_CSUM = (1 << 18),
195 IB_DEVICE_UD_TSO = (1 << 19),
196 IB_DEVICE_XRC = (1 << 20),
199 * This device supports the IB "base memory management extension",
200 * which includes support for fast registrations (IB_WR_REG_MR,
201 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
202 * also be set by any iWarp device which must support FRs to comply
203 * to the iWarp verbs spec. iWarp devices also support the
204 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
207 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
208 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
209 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
210 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
211 IB_DEVICE_RC_IP_CSUM = (1 << 25),
212 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
213 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
215 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
216 * support execution of WQEs that involve synchronization
217 * of I/O operations with single completion queue managed
220 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
221 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
222 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
223 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
224 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
225 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
226 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
227 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
228 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
231 enum ib_signature_prot_cap {
232 IB_PROT_T10DIF_TYPE_1 = 1,
233 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
234 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
237 enum ib_signature_guard_cap {
238 IB_GUARD_T10DIF_CRC = 1,
239 IB_GUARD_T10DIF_CSUM = 1 << 1,
248 enum ib_odp_general_cap_bits {
249 IB_ODP_SUPPORT = 1 << 0,
250 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
253 enum ib_odp_transport_cap_bits {
254 IB_ODP_SUPPORT_SEND = 1 << 0,
255 IB_ODP_SUPPORT_RECV = 1 << 1,
256 IB_ODP_SUPPORT_WRITE = 1 << 2,
257 IB_ODP_SUPPORT_READ = 1 << 3,
258 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
262 uint64_t general_caps;
264 uint32_t rc_odp_caps;
265 uint32_t uc_odp_caps;
266 uint32_t ud_odp_caps;
267 } per_transport_caps;
271 /* Corresponding bit will be set if qp type from
272 * 'enum ib_qp_type' is supported, e.g.
273 * supported_qpts |= 1 << IB_QPT_UD
276 u32 max_rwq_indirection_tables;
277 u32 max_rwq_indirection_table_size;
280 enum ib_cq_creation_flags {
281 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
282 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
285 struct ib_cq_init_attr {
291 struct ib_device_attr {
293 __be64 sys_image_guid;
301 u64 device_cap_flags;
311 int max_qp_init_rd_atom;
312 int max_ee_init_rd_atom;
313 enum ib_atomic_cap atomic_cap;
314 enum ib_atomic_cap masked_atomic_cap;
321 int max_mcast_qp_attach;
322 int max_total_mcast_qp_attach;
329 unsigned int max_fast_reg_page_list_len;
331 u8 local_ca_ack_delay;
334 struct ib_odp_caps odp_caps;
335 uint64_t timestamp_mask;
336 uint64_t hca_core_clock; /* in KHZ */
337 struct ib_rss_caps rss_caps;
339 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
350 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
353 case IB_MTU_256: return 256;
354 case IB_MTU_512: return 512;
355 case IB_MTU_1024: return 1024;
356 case IB_MTU_2048: return 2048;
357 case IB_MTU_4096: return 4096;
362 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
366 else if (mtu >= 2048)
368 else if (mtu >= 1024)
382 IB_PORT_ACTIVE_DEFER = 5
385 enum ib_port_cap_flags {
387 IB_PORT_NOTICE_SUP = 1 << 2,
388 IB_PORT_TRAP_SUP = 1 << 3,
389 IB_PORT_OPT_IPD_SUP = 1 << 4,
390 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
391 IB_PORT_SL_MAP_SUP = 1 << 6,
392 IB_PORT_MKEY_NVRAM = 1 << 7,
393 IB_PORT_PKEY_NVRAM = 1 << 8,
394 IB_PORT_LED_INFO_SUP = 1 << 9,
395 IB_PORT_SM_DISABLED = 1 << 10,
396 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
397 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
398 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
399 IB_PORT_CM_SUP = 1 << 16,
400 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
401 IB_PORT_REINIT_SUP = 1 << 18,
402 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
403 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
404 IB_PORT_DR_NOTICE_SUP = 1 << 21,
405 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
406 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
407 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
408 IB_PORT_CLIENT_REG_SUP = 1 << 25,
409 IB_PORT_IP_BASED_GIDS = 1 << 26,
419 static inline int ib_width_enum_to_int(enum ib_port_width width)
422 case IB_WIDTH_1X: return 1;
423 case IB_WIDTH_4X: return 4;
424 case IB_WIDTH_8X: return 8;
425 case IB_WIDTH_12X: return 12;
441 * struct rdma_hw_stats
442 * @timestamp - Used by the core code to track when the last update was
443 * @lifespan - Used by the core code to determine how old the counters
444 * should be before being updated again. Stored in jiffies, defaults
445 * to 10 milliseconds, drivers can override the default be specifying
446 * their own value during their allocation routine.
447 * @name - Array of pointers to static names used for the counters in
449 * @num_counters - How many hardware counters there are. If name is
450 * shorter than this number, a kernel oops will result. Driver authors
451 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
452 * in their code to prevent this.
453 * @value - Array of u64 counters that are accessed by the sysfs code and
454 * filled in by the drivers get_stats routine
456 struct rdma_hw_stats {
457 unsigned long timestamp;
458 unsigned long lifespan;
459 const char * const *names;
464 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
466 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
468 * @names - Array of static const char *
469 * @num_counters - How many elements in array
470 * @lifespan - How many milliseconds between updates
472 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
473 const char * const *names, int num_counters,
474 unsigned long lifespan)
476 struct rdma_hw_stats *stats;
478 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
482 stats->names = names;
483 stats->num_counters = num_counters;
484 stats->lifespan = msecs_to_jiffies(lifespan);
490 /* Define bits for the various functionality this port needs to be supported by
493 /* Management 0x00000FFF */
494 #define RDMA_CORE_CAP_IB_MAD 0x00000001
495 #define RDMA_CORE_CAP_IB_SMI 0x00000002
496 #define RDMA_CORE_CAP_IB_CM 0x00000004
497 #define RDMA_CORE_CAP_IW_CM 0x00000008
498 #define RDMA_CORE_CAP_IB_SA 0x00000010
499 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
501 /* Address format 0x000FF000 */
502 #define RDMA_CORE_CAP_AF_IB 0x00001000
503 #define RDMA_CORE_CAP_ETH_AH 0x00002000
504 #define RDMA_CORE_CAP_OPA_AH 0x00004000
506 /* Protocol 0xFFF00000 */
507 #define RDMA_CORE_CAP_PROT_IB 0x00100000
508 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
509 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
510 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
511 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
512 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
514 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
515 | RDMA_CORE_CAP_IB_MAD \
516 | RDMA_CORE_CAP_IB_SMI \
517 | RDMA_CORE_CAP_IB_CM \
518 | RDMA_CORE_CAP_IB_SA \
519 | RDMA_CORE_CAP_AF_IB)
520 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
521 | RDMA_CORE_CAP_IB_MAD \
522 | RDMA_CORE_CAP_IB_CM \
523 | RDMA_CORE_CAP_AF_IB \
524 | RDMA_CORE_CAP_ETH_AH)
525 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
526 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
527 | RDMA_CORE_CAP_IB_MAD \
528 | RDMA_CORE_CAP_IB_CM \
529 | RDMA_CORE_CAP_AF_IB \
530 | RDMA_CORE_CAP_ETH_AH)
531 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
532 | RDMA_CORE_CAP_IW_CM)
533 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
534 | RDMA_CORE_CAP_OPA_MAD)
536 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
538 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
540 struct ib_port_attr {
542 enum ib_port_state state;
544 enum ib_mtu active_mtu;
564 enum ib_device_modify_flags {
565 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
566 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
569 #define IB_DEVICE_NODE_DESC_MAX 64
571 struct ib_device_modify {
573 char node_desc[IB_DEVICE_NODE_DESC_MAX];
576 enum ib_port_modify_flags {
577 IB_PORT_SHUTDOWN = 1,
578 IB_PORT_INIT_TYPE = (1<<2),
579 IB_PORT_RESET_QKEY_CNTR = (1<<3)
582 struct ib_port_modify {
583 u32 set_port_cap_mask;
584 u32 clr_port_cap_mask;
592 IB_EVENT_QP_ACCESS_ERR,
596 IB_EVENT_PATH_MIG_ERR,
597 IB_EVENT_DEVICE_FATAL,
598 IB_EVENT_PORT_ACTIVE,
601 IB_EVENT_PKEY_CHANGE,
604 IB_EVENT_SRQ_LIMIT_REACHED,
605 IB_EVENT_QP_LAST_WQE_REACHED,
606 IB_EVENT_CLIENT_REREGISTER,
611 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
614 struct ib_device *device;
622 enum ib_event_type event;
625 struct ib_event_handler {
626 struct ib_device *device;
627 void (*handler)(struct ib_event_handler *, struct ib_event *);
628 struct list_head list;
631 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
633 (_ptr)->device = _device; \
634 (_ptr)->handler = _handler; \
635 INIT_LIST_HEAD(&(_ptr)->list); \
638 struct ib_global_route {
647 __be32 version_tclass_flow;
655 union rdma_network_hdr {
658 /* The IB spec states that if it's IPv4, the header
659 * is located in the last 20 bytes of the header.
662 struct iphdr roce4grh;
667 IB_MULTICAST_QPN = 0xffffff
670 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
671 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
678 IB_RATE_PORT_CURRENT = 0,
679 IB_RATE_2_5_GBPS = 2,
687 IB_RATE_120_GBPS = 10,
688 IB_RATE_14_GBPS = 11,
689 IB_RATE_56_GBPS = 12,
690 IB_RATE_112_GBPS = 13,
691 IB_RATE_168_GBPS = 14,
692 IB_RATE_25_GBPS = 15,
693 IB_RATE_100_GBPS = 16,
694 IB_RATE_200_GBPS = 17,
695 IB_RATE_300_GBPS = 18
699 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
700 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
701 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
702 * @rate: rate to convert.
704 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
707 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
708 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
709 * @rate: rate to convert.
711 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
715 * enum ib_mr_type - memory region type
716 * @IB_MR_TYPE_MEM_REG: memory region that is used for
717 * normal registration
718 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
719 * signature operations (data-integrity
721 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
722 * register any arbitrary sg lists (without
723 * the normal mr constraints - see
728 IB_MR_TYPE_SIGNATURE,
734 * IB_SIG_TYPE_NONE: Unprotected.
735 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
737 enum ib_signature_type {
743 * Signature T10-DIF block-guard types
744 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
745 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
747 enum ib_t10_dif_bg_type {
753 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
755 * @bg_type: T10-DIF block guard type (CRC|CSUM)
756 * @pi_interval: protection information interval.
757 * @bg: seed of guard computation.
758 * @app_tag: application tag of guard block
759 * @ref_tag: initial guard block reference tag.
760 * @ref_remap: Indicate wethear the reftag increments each block
761 * @app_escape: Indicate to skip block check if apptag=0xffff
762 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
763 * @apptag_check_mask: check bitmask of application tag.
765 struct ib_t10_dif_domain {
766 enum ib_t10_dif_bg_type bg_type;
774 u16 apptag_check_mask;
778 * struct ib_sig_domain - Parameters for signature domain
779 * @sig_type: specific signauture type
780 * @sig: union of all signature domain attributes that may
781 * be used to set domain layout.
783 struct ib_sig_domain {
784 enum ib_signature_type sig_type;
786 struct ib_t10_dif_domain dif;
791 * struct ib_sig_attrs - Parameters for signature handover operation
792 * @check_mask: bitmask for signature byte check (8 bytes)
793 * @mem: memory domain layout desciptor.
794 * @wire: wire domain layout desciptor.
796 struct ib_sig_attrs {
798 struct ib_sig_domain mem;
799 struct ib_sig_domain wire;
802 enum ib_sig_err_type {
809 * struct ib_sig_err - signature error descriptor
812 enum ib_sig_err_type err_type;
819 enum ib_mr_status_check {
820 IB_MR_CHECK_SIG_STATUS = 1,
824 * struct ib_mr_status - Memory region status container
826 * @fail_status: Bitmask of MR checks status. For each
827 * failed check a corresponding status bit is set.
828 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
831 struct ib_mr_status {
833 struct ib_sig_err sig_err;
837 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
839 * @mult: multiple to convert.
841 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
843 struct rdma_ah_attr {
844 struct ib_global_route grh;
858 IB_WC_LOC_EEC_OP_ERR,
863 IB_WC_LOC_ACCESS_ERR,
864 IB_WC_REM_INV_REQ_ERR,
865 IB_WC_REM_ACCESS_ERR,
868 IB_WC_RNR_RETRY_EXC_ERR,
869 IB_WC_LOC_RDD_VIOL_ERR,
870 IB_WC_REM_INV_RD_REQ_ERR,
873 IB_WC_INV_EEC_STATE_ERR,
875 IB_WC_RESP_TIMEOUT_ERR,
879 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
890 IB_WC_MASKED_COMP_SWAP,
891 IB_WC_MASKED_FETCH_ADD,
893 * Set value of IB_WC_RECV so consumers can test if a completion is a
894 * receive by testing (opcode & IB_WC_RECV).
897 IB_WC_RECV_RDMA_WITH_IMM
902 IB_WC_WITH_IMM = (1<<1),
903 IB_WC_WITH_INVALIDATE = (1<<2),
904 IB_WC_IP_CSUM_OK = (1<<3),
905 IB_WC_WITH_SMAC = (1<<4),
906 IB_WC_WITH_VLAN = (1<<5),
907 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
913 struct ib_cqe *wr_cqe;
915 enum ib_wc_status status;
916 enum ib_wc_opcode opcode;
930 u8 port_num; /* valid only for DR SMPs on switches */
936 enum ib_cq_notify_flags {
937 IB_CQ_SOLICITED = 1 << 0,
938 IB_CQ_NEXT_COMP = 1 << 1,
939 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
940 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
948 enum ib_srq_attr_mask {
949 IB_SRQ_MAX_WR = 1 << 0,
950 IB_SRQ_LIMIT = 1 << 1,
959 struct ib_srq_init_attr {
960 void (*event_handler)(struct ib_event *, void *);
962 struct ib_srq_attr attr;
963 enum ib_srq_type srq_type;
967 struct ib_xrcd *xrcd;
981 * Maximum number of rdma_rw_ctx structures in flight at a time.
982 * ib_create_qp() will calculate the right amount of neededed WRs
983 * and MRs based on this.
995 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
996 * here (and in that order) since the MAD layer uses them as
997 * indices into a 2-entry table.
1006 IB_QPT_RAW_ETHERTYPE,
1007 IB_QPT_RAW_PACKET = 8,
1011 /* Reserve a range for qp types internal to the low level driver.
1012 * These qp types will not be visible at the IB core layer, so the
1013 * IB_QPT_MAX usages should not be affected in the core layer
1015 IB_QPT_RESERVED1 = 0x1000,
1027 enum ib_qp_create_flags {
1028 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1029 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1030 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1031 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1032 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1033 IB_QP_CREATE_NETIF_QP = 1 << 5,
1034 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1035 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
1036 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1037 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1038 /* reserve bits 26-31 for low level drivers' internal use */
1039 IB_QP_CREATE_RESERVED_START = 1 << 26,
1040 IB_QP_CREATE_RESERVED_END = 1 << 31,
1044 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1045 * callback to destroy the passed in QP.
1048 struct ib_qp_init_attr {
1049 void (*event_handler)(struct ib_event *, void *);
1051 struct ib_cq *send_cq;
1052 struct ib_cq *recv_cq;
1054 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1055 struct ib_qp_cap cap;
1056 enum ib_sig_type sq_sig_type;
1057 enum ib_qp_type qp_type;
1058 enum ib_qp_create_flags create_flags;
1061 * Only needed for special QP types, or when using the RW API.
1064 struct ib_rwq_ind_table *rwq_ind_tbl;
1067 struct ib_qp_open_attr {
1068 void (*event_handler)(struct ib_event *, void *);
1071 enum ib_qp_type qp_type;
1074 enum ib_rnr_timeout {
1075 IB_RNR_TIMER_655_36 = 0,
1076 IB_RNR_TIMER_000_01 = 1,
1077 IB_RNR_TIMER_000_02 = 2,
1078 IB_RNR_TIMER_000_03 = 3,
1079 IB_RNR_TIMER_000_04 = 4,
1080 IB_RNR_TIMER_000_06 = 5,
1081 IB_RNR_TIMER_000_08 = 6,
1082 IB_RNR_TIMER_000_12 = 7,
1083 IB_RNR_TIMER_000_16 = 8,
1084 IB_RNR_TIMER_000_24 = 9,
1085 IB_RNR_TIMER_000_32 = 10,
1086 IB_RNR_TIMER_000_48 = 11,
1087 IB_RNR_TIMER_000_64 = 12,
1088 IB_RNR_TIMER_000_96 = 13,
1089 IB_RNR_TIMER_001_28 = 14,
1090 IB_RNR_TIMER_001_92 = 15,
1091 IB_RNR_TIMER_002_56 = 16,
1092 IB_RNR_TIMER_003_84 = 17,
1093 IB_RNR_TIMER_005_12 = 18,
1094 IB_RNR_TIMER_007_68 = 19,
1095 IB_RNR_TIMER_010_24 = 20,
1096 IB_RNR_TIMER_015_36 = 21,
1097 IB_RNR_TIMER_020_48 = 22,
1098 IB_RNR_TIMER_030_72 = 23,
1099 IB_RNR_TIMER_040_96 = 24,
1100 IB_RNR_TIMER_061_44 = 25,
1101 IB_RNR_TIMER_081_92 = 26,
1102 IB_RNR_TIMER_122_88 = 27,
1103 IB_RNR_TIMER_163_84 = 28,
1104 IB_RNR_TIMER_245_76 = 29,
1105 IB_RNR_TIMER_327_68 = 30,
1106 IB_RNR_TIMER_491_52 = 31
1109 enum ib_qp_attr_mask {
1111 IB_QP_CUR_STATE = (1<<1),
1112 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1113 IB_QP_ACCESS_FLAGS = (1<<3),
1114 IB_QP_PKEY_INDEX = (1<<4),
1115 IB_QP_PORT = (1<<5),
1116 IB_QP_QKEY = (1<<6),
1118 IB_QP_PATH_MTU = (1<<8),
1119 IB_QP_TIMEOUT = (1<<9),
1120 IB_QP_RETRY_CNT = (1<<10),
1121 IB_QP_RNR_RETRY = (1<<11),
1122 IB_QP_RQ_PSN = (1<<12),
1123 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1124 IB_QP_ALT_PATH = (1<<14),
1125 IB_QP_MIN_RNR_TIMER = (1<<15),
1126 IB_QP_SQ_PSN = (1<<16),
1127 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1128 IB_QP_PATH_MIG_STATE = (1<<18),
1129 IB_QP_CAP = (1<<19),
1130 IB_QP_DEST_QPN = (1<<20),
1131 IB_QP_RESERVED1 = (1<<21),
1132 IB_QP_RESERVED2 = (1<<22),
1133 IB_QP_RESERVED3 = (1<<23),
1134 IB_QP_RESERVED4 = (1<<24),
1135 IB_QP_RATE_LIMIT = (1<<25),
1160 enum ib_qp_state qp_state;
1161 enum ib_qp_state cur_qp_state;
1162 enum ib_mtu path_mtu;
1163 enum ib_mig_state path_mig_state;
1168 int qp_access_flags;
1169 struct ib_qp_cap cap;
1170 struct rdma_ah_attr ah_attr;
1171 struct rdma_ah_attr alt_ah_attr;
1174 u8 en_sqd_async_notify;
1177 u8 max_dest_rd_atomic;
1190 IB_WR_RDMA_WRITE_WITH_IMM,
1192 IB_WR_SEND_WITH_IMM,
1194 IB_WR_ATOMIC_CMP_AND_SWP,
1195 IB_WR_ATOMIC_FETCH_AND_ADD,
1197 IB_WR_SEND_WITH_INV,
1198 IB_WR_RDMA_READ_WITH_INV,
1201 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1202 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1204 /* reserve values for low level drivers' internal use.
1205 * These values will not be used at all in the ib core layer.
1207 IB_WR_RESERVED1 = 0xf0,
1219 enum ib_send_flags {
1221 IB_SEND_SIGNALED = (1<<1),
1222 IB_SEND_SOLICITED = (1<<2),
1223 IB_SEND_INLINE = (1<<3),
1224 IB_SEND_IP_CSUM = (1<<4),
1226 /* reserve bits 26-31 for low level drivers' internal use */
1227 IB_SEND_RESERVED_START = (1 << 26),
1228 IB_SEND_RESERVED_END = (1 << 31),
1238 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1242 struct ib_send_wr *next;
1245 struct ib_cqe *wr_cqe;
1247 struct ib_sge *sg_list;
1249 enum ib_wr_opcode opcode;
1253 u32 invalidate_rkey;
1258 struct ib_send_wr wr;
1263 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1265 return container_of(wr, struct ib_rdma_wr, wr);
1268 struct ib_atomic_wr {
1269 struct ib_send_wr wr;
1273 u64 compare_add_mask;
1278 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1280 return container_of(wr, struct ib_atomic_wr, wr);
1284 struct ib_send_wr wr;
1291 u16 pkey_index; /* valid for GSI only */
1292 u8 port_num; /* valid for DR SMPs on switch only */
1295 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1297 return container_of(wr, struct ib_ud_wr, wr);
1301 struct ib_send_wr wr;
1307 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1309 return container_of(wr, struct ib_reg_wr, wr);
1312 struct ib_sig_handover_wr {
1313 struct ib_send_wr wr;
1314 struct ib_sig_attrs *sig_attrs;
1315 struct ib_mr *sig_mr;
1317 struct ib_sge *prot;
1320 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1322 return container_of(wr, struct ib_sig_handover_wr, wr);
1326 struct ib_recv_wr *next;
1329 struct ib_cqe *wr_cqe;
1331 struct ib_sge *sg_list;
1335 enum ib_access_flags {
1336 IB_ACCESS_LOCAL_WRITE = 1,
1337 IB_ACCESS_REMOTE_WRITE = (1<<1),
1338 IB_ACCESS_REMOTE_READ = (1<<2),
1339 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1340 IB_ACCESS_MW_BIND = (1<<4),
1341 IB_ZERO_BASED = (1<<5),
1342 IB_ACCESS_ON_DEMAND = (1<<6),
1343 IB_ACCESS_HUGETLB = (1<<7),
1347 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1348 * are hidden here instead of a uapi header!
1350 enum ib_mr_rereg_flags {
1351 IB_MR_REREG_TRANS = 1,
1352 IB_MR_REREG_PD = (1<<1),
1353 IB_MR_REREG_ACCESS = (1<<2),
1354 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1357 struct ib_fmr_attr {
1365 enum rdma_remove_reason {
1366 /* Userspace requested uobject deletion. Call could fail */
1367 RDMA_REMOVE_DESTROY,
1368 /* Context deletion. This call should delete the actual object itself */
1370 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1371 RDMA_REMOVE_DRIVER_REMOVE,
1372 /* Context is being cleaned-up, but commit was just completed */
1373 RDMA_REMOVE_DURING_CLEANUP,
1376 struct ib_rdmacg_object {
1377 #ifdef CONFIG_CGROUP_RDMA
1378 struct rdma_cgroup *cg; /* owner rdma cgroup */
1382 struct ib_ucontext {
1383 struct ib_device *device;
1384 struct ib_uverbs_file *ufile;
1387 /* locking the uobjects_list */
1388 struct mutex uobjects_lock;
1389 struct list_head uobjects;
1390 /* protects cleanup process from other actions */
1391 struct rw_semaphore cleanup_rwsem;
1392 enum rdma_remove_reason cleanup_reason;
1395 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1396 struct rb_root umem_tree;
1398 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1399 * mmu notifiers registration.
1401 struct rw_semaphore umem_rwsem;
1402 void (*invalidate_range)(struct ib_umem *umem,
1403 unsigned long start, unsigned long end);
1405 struct mmu_notifier mn;
1406 atomic_t notifier_count;
1407 /* A list of umems that don't have private mmu notifier counters yet. */
1408 struct list_head no_private_counters;
1412 struct ib_rdmacg_object cg_obj;
1416 u64 user_handle; /* handle given to us by userspace */
1417 struct ib_ucontext *context; /* associated user context */
1418 void *object; /* containing object */
1419 struct list_head list; /* link to context's list */
1420 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1421 int id; /* index into kernel idr */
1423 atomic_t usecnt; /* protects exclusive access */
1424 struct rcu_head rcu; /* kfree_rcu() overhead */
1426 const struct uverbs_obj_type *type;
1429 struct ib_uobject_file {
1430 struct ib_uobject uobj;
1431 /* ufile contains the lock between context release and file close */
1432 struct ib_uverbs_file *ufile;
1436 const void __user *inbuf;
1437 void __user *outbuf;
1445 struct ib_device *device;
1446 struct ib_uobject *uobject;
1447 atomic_t usecnt; /* count all resources */
1449 u32 unsafe_global_rkey;
1452 * Implementation details of the RDMA core, don't use in drivers:
1454 struct ib_mr *__internal_mr;
1458 struct ib_device *device;
1459 atomic_t usecnt; /* count all exposed resources */
1460 struct inode *inode;
1462 struct mutex tgt_qp_mutex;
1463 struct list_head tgt_qp_list;
1467 struct ib_device *device;
1469 struct ib_uobject *uobject;
1472 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1474 enum ib_poll_context {
1475 IB_POLL_DIRECT, /* caller context, no hw completions */
1476 IB_POLL_SOFTIRQ, /* poll from softirq context */
1477 IB_POLL_WORKQUEUE, /* poll from workqueue */
1481 struct ib_device *device;
1482 struct ib_uobject *uobject;
1483 ib_comp_handler comp_handler;
1484 void (*event_handler)(struct ib_event *, void *);
1487 atomic_t usecnt; /* count number of work queues */
1488 enum ib_poll_context poll_ctx;
1491 struct irq_poll iop;
1492 struct work_struct work;
1497 struct ib_device *device;
1499 struct ib_uobject *uobject;
1500 void (*event_handler)(struct ib_event *, void *);
1502 enum ib_srq_type srq_type;
1507 struct ib_xrcd *xrcd;
1514 enum ib_raw_packet_caps {
1515 /* Strip cvlan from incoming packet and report it in the matching work
1516 * completion is supported.
1518 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1519 /* Scatter FCS field of an incoming packet to host memory is supported.
1521 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1522 /* Checksum offloads are supported (for both send and receive). */
1523 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1537 struct ib_device *device;
1538 struct ib_uobject *uobject;
1540 void (*event_handler)(struct ib_event *, void *);
1544 enum ib_wq_state state;
1545 enum ib_wq_type wq_type;
1550 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1551 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1554 struct ib_wq_init_attr {
1556 enum ib_wq_type wq_type;
1560 void (*event_handler)(struct ib_event *, void *);
1561 u32 create_flags; /* Use enum ib_wq_flags */
1564 enum ib_wq_attr_mask {
1565 IB_WQ_STATE = 1 << 0,
1566 IB_WQ_CUR_STATE = 1 << 1,
1567 IB_WQ_FLAGS = 1 << 2,
1571 enum ib_wq_state wq_state;
1572 enum ib_wq_state curr_wq_state;
1573 u32 flags; /* Use enum ib_wq_flags */
1574 u32 flags_mask; /* Use enum ib_wq_flags */
1577 struct ib_rwq_ind_table {
1578 struct ib_device *device;
1579 struct ib_uobject *uobject;
1582 u32 log_ind_tbl_size;
1583 struct ib_wq **ind_tbl;
1586 struct ib_rwq_ind_table_init_attr {
1587 u32 log_ind_tbl_size;
1588 /* Each entry is a pointer to Receive Work Queue */
1589 struct ib_wq **ind_tbl;
1593 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1594 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1597 struct ib_device *device;
1599 struct ib_cq *send_cq;
1600 struct ib_cq *recv_cq;
1603 struct list_head rdma_mrs;
1604 struct list_head sig_mrs;
1606 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1607 struct list_head xrcd_list;
1609 /* count times opened, mcast attaches, flow attaches */
1611 struct list_head open_list;
1612 struct ib_qp *real_qp;
1613 struct ib_uobject *uobject;
1614 void (*event_handler)(struct ib_event *, void *);
1619 enum ib_qp_type qp_type;
1620 struct ib_rwq_ind_table *rwq_ind_tbl;
1624 struct ib_device *device;
1630 unsigned int page_size;
1633 struct ib_uobject *uobject; /* user */
1634 struct list_head qp_entry; /* FR */
1639 struct ib_device *device;
1641 struct ib_uobject *uobject;
1643 enum ib_mw_type type;
1647 struct ib_device *device;
1649 struct list_head list;
1654 /* Supported steering options */
1655 enum ib_flow_attr_type {
1656 /* steering according to rule specifications */
1657 IB_FLOW_ATTR_NORMAL = 0x0,
1658 /* default unicast and multicast rule -
1659 * receive all Eth traffic which isn't steered to any QP
1661 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1662 /* default multicast rule -
1663 * receive all Eth multicast traffic which isn't steered to any QP
1665 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1666 /* sniffer rule - receive all port traffic */
1667 IB_FLOW_ATTR_SNIFFER = 0x3
1670 /* Supported steering header types */
1671 enum ib_flow_spec_type {
1673 IB_FLOW_SPEC_ETH = 0x20,
1674 IB_FLOW_SPEC_IB = 0x22,
1676 IB_FLOW_SPEC_IPV4 = 0x30,
1677 IB_FLOW_SPEC_IPV6 = 0x31,
1679 IB_FLOW_SPEC_TCP = 0x40,
1680 IB_FLOW_SPEC_UDP = 0x41,
1681 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1682 IB_FLOW_SPEC_INNER = 0x100,
1684 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1685 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1687 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1688 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1690 /* Flow steering rule priority is set according to it's domain.
1691 * Lower domain value means higher priority.
1693 enum ib_flow_domain {
1694 IB_FLOW_DOMAIN_USER,
1695 IB_FLOW_DOMAIN_ETHTOOL,
1698 IB_FLOW_DOMAIN_NUM /* Must be last */
1701 enum ib_flow_flags {
1702 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1703 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1706 struct ib_flow_eth_filter {
1715 struct ib_flow_spec_eth {
1718 struct ib_flow_eth_filter val;
1719 struct ib_flow_eth_filter mask;
1722 struct ib_flow_ib_filter {
1729 struct ib_flow_spec_ib {
1732 struct ib_flow_ib_filter val;
1733 struct ib_flow_ib_filter mask;
1736 /* IPv4 header flags */
1737 enum ib_ipv4_flags {
1738 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1739 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1740 last have this flag set */
1743 struct ib_flow_ipv4_filter {
1754 struct ib_flow_spec_ipv4 {
1757 struct ib_flow_ipv4_filter val;
1758 struct ib_flow_ipv4_filter mask;
1761 struct ib_flow_ipv6_filter {
1772 struct ib_flow_spec_ipv6 {
1775 struct ib_flow_ipv6_filter val;
1776 struct ib_flow_ipv6_filter mask;
1779 struct ib_flow_tcp_udp_filter {
1786 struct ib_flow_spec_tcp_udp {
1789 struct ib_flow_tcp_udp_filter val;
1790 struct ib_flow_tcp_udp_filter mask;
1793 struct ib_flow_tunnel_filter {
1798 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1799 * the tunnel_id from val has the vni value
1801 struct ib_flow_spec_tunnel {
1804 struct ib_flow_tunnel_filter val;
1805 struct ib_flow_tunnel_filter mask;
1808 struct ib_flow_spec_action_tag {
1809 enum ib_flow_spec_type type;
1814 struct ib_flow_spec_action_drop {
1815 enum ib_flow_spec_type type;
1819 union ib_flow_spec {
1824 struct ib_flow_spec_eth eth;
1825 struct ib_flow_spec_ib ib;
1826 struct ib_flow_spec_ipv4 ipv4;
1827 struct ib_flow_spec_tcp_udp tcp_udp;
1828 struct ib_flow_spec_ipv6 ipv6;
1829 struct ib_flow_spec_tunnel tunnel;
1830 struct ib_flow_spec_action_tag flow_tag;
1831 struct ib_flow_spec_action_drop drop;
1834 struct ib_flow_attr {
1835 enum ib_flow_attr_type type;
1841 /* Following are the optional layers according to user request
1842 * struct ib_flow_spec_xxx
1843 * struct ib_flow_spec_yyy
1849 struct ib_uobject *uobject;
1855 enum ib_process_mad_flags {
1856 IB_MAD_IGNORE_MKEY = 1,
1857 IB_MAD_IGNORE_BKEY = 2,
1858 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1861 enum ib_mad_result {
1862 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1863 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1864 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1865 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1868 #define IB_DEVICE_NAME_MAX 64
1870 struct ib_port_cache {
1871 struct ib_pkey_cache *pkey;
1872 struct ib_gid_table *gid;
1874 enum ib_port_state port_state;
1879 struct ib_event_handler event_handler;
1880 struct ib_port_cache *ports;
1885 struct ib_port_immutable {
1892 /* rdma netdev type - specifies protocol type */
1893 enum rdma_netdev_t {
1894 RDMA_NETDEV_OPA_VNIC,
1899 * struct rdma_netdev - rdma netdev
1900 * For cases where netstack interfacing is required.
1902 struct rdma_netdev {
1904 struct ib_device *hca;
1907 /* control functions */
1908 void (*set_id)(struct net_device *netdev, int id);
1910 int (*send)(struct net_device *dev, struct sk_buff *skb,
1911 struct ib_ah *address, u32 dqpn);
1913 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
1914 union ib_gid *gid, u16 mlid,
1915 int set_qkey, u32 qkey);
1916 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
1917 union ib_gid *gid, u16 mlid);
1921 /* Do not access @dma_device directly from ULP nor from HW drivers. */
1922 struct device *dma_device;
1924 char name[IB_DEVICE_NAME_MAX];
1926 struct list_head event_handler_list;
1927 spinlock_t event_handler_lock;
1929 spinlock_t client_data_lock;
1930 struct list_head core_list;
1931 /* Access to the client_data_list is protected by the client_data_lock
1932 * spinlock and the lists_rwsem read-write semaphore */
1933 struct list_head client_data_list;
1935 struct ib_cache cache;
1937 * port_immutable is indexed by port number
1939 struct ib_port_immutable *port_immutable;
1941 int num_comp_vectors;
1943 struct iw_cm_verbs *iwcm;
1946 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1947 * driver initialized data. The struct is kfree()'ed by the sysfs
1948 * core when the device is removed. A lifespan of -1 in the return
1949 * struct tells the core to set a default lifespan.
1951 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1954 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1955 * @index - The index in the value array we wish to have updated, or
1956 * num_counters if we want all stats updated
1958 * < 0 - Error, no counters updated
1959 * index - Updated the single counter pointed to by index
1960 * num_counters - Updated all counters (will reset the timestamp
1961 * and prevent further calls for lifespan milliseconds)
1962 * Drivers are allowed to update all counters in leiu of just the
1963 * one given in index at their option
1965 int (*get_hw_stats)(struct ib_device *device,
1966 struct rdma_hw_stats *stats,
1967 u8 port, int index);
1968 int (*query_device)(struct ib_device *device,
1969 struct ib_device_attr *device_attr,
1970 struct ib_udata *udata);
1971 int (*query_port)(struct ib_device *device,
1973 struct ib_port_attr *port_attr);
1974 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1976 /* When calling get_netdev, the HW vendor's driver should return the
1977 * net device of device @device at port @port_num or NULL if such
1978 * a net device doesn't exist. The vendor driver should call dev_hold
1979 * on this net device. The HW vendor's device driver must guarantee
1980 * that this function returns NULL before the net device reaches
1981 * NETDEV_UNREGISTER_FINAL state.
1983 struct net_device *(*get_netdev)(struct ib_device *device,
1985 int (*query_gid)(struct ib_device *device,
1986 u8 port_num, int index,
1988 /* When calling add_gid, the HW vendor's driver should
1989 * add the gid of device @device at gid index @index of
1990 * port @port_num to be @gid. Meta-info of that gid (for example,
1991 * the network device related to this gid is available
1992 * at @attr. @context allows the HW vendor driver to store extra
1993 * information together with a GID entry. The HW vendor may allocate
1994 * memory to contain this information and store it in @context when a
1995 * new GID entry is written to. Params are consistent until the next
1996 * call of add_gid or delete_gid. The function should return 0 on
1997 * success or error otherwise. The function could be called
1998 * concurrently for different ports. This function is only called
1999 * when roce_gid_table is used.
2001 int (*add_gid)(struct ib_device *device,
2004 const union ib_gid *gid,
2005 const struct ib_gid_attr *attr,
2007 /* When calling del_gid, the HW vendor's driver should delete the
2008 * gid of device @device at gid index @index of port @port_num.
2009 * Upon the deletion of a GID entry, the HW vendor must free any
2010 * allocated memory. The caller will clear @context afterwards.
2011 * This function is only called when roce_gid_table is used.
2013 int (*del_gid)(struct ib_device *device,
2017 int (*query_pkey)(struct ib_device *device,
2018 u8 port_num, u16 index, u16 *pkey);
2019 int (*modify_device)(struct ib_device *device,
2020 int device_modify_mask,
2021 struct ib_device_modify *device_modify);
2022 int (*modify_port)(struct ib_device *device,
2023 u8 port_num, int port_modify_mask,
2024 struct ib_port_modify *port_modify);
2025 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2026 struct ib_udata *udata);
2027 int (*dealloc_ucontext)(struct ib_ucontext *context);
2028 int (*mmap)(struct ib_ucontext *context,
2029 struct vm_area_struct *vma);
2030 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2031 struct ib_ucontext *context,
2032 struct ib_udata *udata);
2033 int (*dealloc_pd)(struct ib_pd *pd);
2034 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2035 struct rdma_ah_attr *ah_attr,
2036 struct ib_udata *udata);
2037 int (*modify_ah)(struct ib_ah *ah,
2038 struct rdma_ah_attr *ah_attr);
2039 int (*query_ah)(struct ib_ah *ah,
2040 struct rdma_ah_attr *ah_attr);
2041 int (*destroy_ah)(struct ib_ah *ah);
2042 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2043 struct ib_srq_init_attr *srq_init_attr,
2044 struct ib_udata *udata);
2045 int (*modify_srq)(struct ib_srq *srq,
2046 struct ib_srq_attr *srq_attr,
2047 enum ib_srq_attr_mask srq_attr_mask,
2048 struct ib_udata *udata);
2049 int (*query_srq)(struct ib_srq *srq,
2050 struct ib_srq_attr *srq_attr);
2051 int (*destroy_srq)(struct ib_srq *srq);
2052 int (*post_srq_recv)(struct ib_srq *srq,
2053 struct ib_recv_wr *recv_wr,
2054 struct ib_recv_wr **bad_recv_wr);
2055 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2056 struct ib_qp_init_attr *qp_init_attr,
2057 struct ib_udata *udata);
2058 int (*modify_qp)(struct ib_qp *qp,
2059 struct ib_qp_attr *qp_attr,
2061 struct ib_udata *udata);
2062 int (*query_qp)(struct ib_qp *qp,
2063 struct ib_qp_attr *qp_attr,
2065 struct ib_qp_init_attr *qp_init_attr);
2066 int (*destroy_qp)(struct ib_qp *qp);
2067 int (*post_send)(struct ib_qp *qp,
2068 struct ib_send_wr *send_wr,
2069 struct ib_send_wr **bad_send_wr);
2070 int (*post_recv)(struct ib_qp *qp,
2071 struct ib_recv_wr *recv_wr,
2072 struct ib_recv_wr **bad_recv_wr);
2073 struct ib_cq * (*create_cq)(struct ib_device *device,
2074 const struct ib_cq_init_attr *attr,
2075 struct ib_ucontext *context,
2076 struct ib_udata *udata);
2077 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2079 int (*destroy_cq)(struct ib_cq *cq);
2080 int (*resize_cq)(struct ib_cq *cq, int cqe,
2081 struct ib_udata *udata);
2082 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2084 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2085 int (*req_notify_cq)(struct ib_cq *cq,
2086 enum ib_cq_notify_flags flags);
2087 int (*req_ncomp_notif)(struct ib_cq *cq,
2089 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2090 int mr_access_flags);
2091 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2092 u64 start, u64 length,
2094 int mr_access_flags,
2095 struct ib_udata *udata);
2096 int (*rereg_user_mr)(struct ib_mr *mr,
2098 u64 start, u64 length,
2100 int mr_access_flags,
2102 struct ib_udata *udata);
2103 int (*dereg_mr)(struct ib_mr *mr);
2104 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2105 enum ib_mr_type mr_type,
2107 int (*map_mr_sg)(struct ib_mr *mr,
2108 struct scatterlist *sg,
2110 unsigned int *sg_offset);
2111 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2112 enum ib_mw_type type,
2113 struct ib_udata *udata);
2114 int (*dealloc_mw)(struct ib_mw *mw);
2115 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2116 int mr_access_flags,
2117 struct ib_fmr_attr *fmr_attr);
2118 int (*map_phys_fmr)(struct ib_fmr *fmr,
2119 u64 *page_list, int list_len,
2121 int (*unmap_fmr)(struct list_head *fmr_list);
2122 int (*dealloc_fmr)(struct ib_fmr *fmr);
2123 int (*attach_mcast)(struct ib_qp *qp,
2126 int (*detach_mcast)(struct ib_qp *qp,
2129 int (*process_mad)(struct ib_device *device,
2130 int process_mad_flags,
2132 const struct ib_wc *in_wc,
2133 const struct ib_grh *in_grh,
2134 const struct ib_mad_hdr *in_mad,
2136 struct ib_mad_hdr *out_mad,
2137 size_t *out_mad_size,
2138 u16 *out_mad_pkey_index);
2139 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2140 struct ib_ucontext *ucontext,
2141 struct ib_udata *udata);
2142 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2143 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2147 int (*destroy_flow)(struct ib_flow *flow_id);
2148 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2149 struct ib_mr_status *mr_status);
2150 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2151 void (*drain_rq)(struct ib_qp *qp);
2152 void (*drain_sq)(struct ib_qp *qp);
2153 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2155 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2156 struct ifla_vf_info *ivf);
2157 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2158 struct ifla_vf_stats *stats);
2159 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2161 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2162 struct ib_wq_init_attr *init_attr,
2163 struct ib_udata *udata);
2164 int (*destroy_wq)(struct ib_wq *wq);
2165 int (*modify_wq)(struct ib_wq *wq,
2166 struct ib_wq_attr *attr,
2168 struct ib_udata *udata);
2169 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2170 struct ib_rwq_ind_table_init_attr *init_attr,
2171 struct ib_udata *udata);
2172 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2174 * rdma netdev operations
2176 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2177 * doesn't support the specified rdma netdev type.
2179 struct net_device *(*alloc_rdma_netdev)(
2180 struct ib_device *device,
2182 enum rdma_netdev_t type,
2184 unsigned char name_assign_type,
2185 void (*setup)(struct net_device *));
2186 void (*free_rdma_netdev)(struct net_device *netdev);
2188 struct module *owner;
2190 struct kobject *ports_parent;
2191 struct list_head port_list;
2194 IB_DEV_UNINITIALIZED,
2200 u64 uverbs_cmd_mask;
2201 u64 uverbs_ex_cmd_mask;
2203 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2209 struct ib_device_attr attrs;
2210 struct attribute_group *hw_stats_ag;
2211 struct rdma_hw_stats *hw_stats;
2213 #ifdef CONFIG_CGROUP_RDMA
2214 struct rdmacg_device cg_device;
2218 * The following mandatory functions are used only at device
2219 * registration. Keep functions such as these at the end of this
2220 * structure to avoid cache line misses when accessing struct ib_device
2223 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2224 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2229 void (*add) (struct ib_device *);
2230 void (*remove)(struct ib_device *, void *client_data);
2232 /* Returns the net_dev belonging to this ib_client and matching the
2234 * @dev: An RDMA device that the net_dev use for communication.
2235 * @port: A physical port number on the RDMA device.
2236 * @pkey: P_Key that the net_dev uses if applicable.
2237 * @gid: A GID that the net_dev uses to communicate.
2238 * @addr: An IP address the net_dev is configured with.
2239 * @client_data: The device's client data set by ib_set_client_data().
2241 * An ib_client that implements a net_dev on top of RDMA devices
2242 * (such as IP over IB) should implement this callback, allowing the
2243 * rdma_cm module to find the right net_dev for a given request.
2245 * The caller is responsible for calling dev_put on the returned
2247 struct net_device *(*get_net_dev_by_params)(
2248 struct ib_device *dev,
2251 const union ib_gid *gid,
2252 const struct sockaddr *addr,
2254 struct list_head list;
2257 struct ib_device *ib_alloc_device(size_t size);
2258 void ib_dealloc_device(struct ib_device *device);
2260 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2262 int ib_register_device(struct ib_device *device,
2263 int (*port_callback)(struct ib_device *,
2264 u8, struct kobject *));
2265 void ib_unregister_device(struct ib_device *device);
2267 int ib_register_client (struct ib_client *client);
2268 void ib_unregister_client(struct ib_client *client);
2270 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2271 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2274 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2276 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2279 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2281 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2284 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2288 const void __user *p = udata->inbuf + offset;
2292 if (len > USHRT_MAX)
2295 buf = memdup_user(p, len);
2299 ret = !memchr_inv(buf, 0, len);
2305 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2306 * contains all required attributes and no attributes not allowed for
2307 * the given QP state transition.
2308 * @cur_state: Current QP state
2309 * @next_state: Next QP state
2311 * @mask: Mask of supplied QP attributes
2312 * @ll : link layer of port
2314 * This function is a helper function that a low-level driver's
2315 * modify_qp method can use to validate the consumer's input. It
2316 * checks that cur_state and next_state are valid QP states, that a
2317 * transition from cur_state to next_state is allowed by the IB spec,
2318 * and that the attribute mask supplied is allowed for the transition.
2320 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2321 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2322 enum rdma_link_layer ll);
2324 int ib_register_event_handler (struct ib_event_handler *event_handler);
2325 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2326 void ib_dispatch_event(struct ib_event *event);
2328 int ib_query_port(struct ib_device *device,
2329 u8 port_num, struct ib_port_attr *port_attr);
2331 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2335 * rdma_cap_ib_switch - Check if the device is IB switch
2336 * @device: Device to check
2338 * Device driver is responsible for setting is_switch bit on
2339 * in ib_device structure at init time.
2341 * Return: true if the device is IB switch.
2343 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2345 return device->is_switch;
2349 * rdma_start_port - Return the first valid port number for the device
2352 * @device: Device to be checked
2354 * Return start port number
2356 static inline u8 rdma_start_port(const struct ib_device *device)
2358 return rdma_cap_ib_switch(device) ? 0 : 1;
2362 * rdma_end_port - Return the last valid port number for the device
2365 * @device: Device to be checked
2367 * Return last port number
2369 static inline u8 rdma_end_port(const struct ib_device *device)
2371 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2374 static inline int rdma_is_port_valid(const struct ib_device *device,
2377 return (port >= rdma_start_port(device) &&
2378 port <= rdma_end_port(device));
2381 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2383 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2386 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2388 return device->port_immutable[port_num].core_cap_flags &
2389 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2392 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2394 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2397 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2399 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2402 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2404 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2407 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2409 return rdma_protocol_ib(device, port_num) ||
2410 rdma_protocol_roce(device, port_num);
2413 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2415 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2418 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2420 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2424 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2425 * Management Datagrams.
2426 * @device: Device to check
2427 * @port_num: Port number to check
2429 * Management Datagrams (MAD) are a required part of the InfiniBand
2430 * specification and are supported on all InfiniBand devices. A slightly
2431 * extended version are also supported on OPA interfaces.
2433 * Return: true if the port supports sending/receiving of MAD packets.
2435 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2437 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2441 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2442 * Management Datagrams.
2443 * @device: Device to check
2444 * @port_num: Port number to check
2446 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2447 * datagrams with their own versions. These OPA MADs share many but not all of
2448 * the characteristics of InfiniBand MADs.
2450 * OPA MADs differ in the following ways:
2452 * 1) MADs are variable size up to 2K
2453 * IBTA defined MADs remain fixed at 256 bytes
2454 * 2) OPA SMPs must carry valid PKeys
2455 * 3) OPA SMP packets are a different format
2457 * Return: true if the port supports OPA MAD packet formats.
2459 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2461 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2462 == RDMA_CORE_CAP_OPA_MAD;
2466 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2467 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2468 * @device: Device to check
2469 * @port_num: Port number to check
2471 * Each InfiniBand node is required to provide a Subnet Management Agent
2472 * that the subnet manager can access. Prior to the fabric being fully
2473 * configured by the subnet manager, the SMA is accessed via a well known
2474 * interface called the Subnet Management Interface (SMI). This interface
2475 * uses directed route packets to communicate with the SM to get around the
2476 * chicken and egg problem of the SM needing to know what's on the fabric
2477 * in order to configure the fabric, and needing to configure the fabric in
2478 * order to send packets to the devices on the fabric. These directed
2479 * route packets do not need the fabric fully configured in order to reach
2480 * their destination. The SMI is the only method allowed to send
2481 * directed route packets on an InfiniBand fabric.
2483 * Return: true if the port provides an SMI.
2485 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2487 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2491 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2492 * Communication Manager.
2493 * @device: Device to check
2494 * @port_num: Port number to check
2496 * The InfiniBand Communication Manager is one of many pre-defined General
2497 * Service Agents (GSA) that are accessed via the General Service
2498 * Interface (GSI). It's role is to facilitate establishment of connections
2499 * between nodes as well as other management related tasks for established
2502 * Return: true if the port supports an IB CM (this does not guarantee that
2503 * a CM is actually running however).
2505 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2507 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2511 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2512 * Communication Manager.
2513 * @device: Device to check
2514 * @port_num: Port number to check
2516 * Similar to above, but specific to iWARP connections which have a different
2517 * managment protocol than InfiniBand.
2519 * Return: true if the port supports an iWARP CM (this does not guarantee that
2520 * a CM is actually running however).
2522 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2524 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2528 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2529 * Subnet Administration.
2530 * @device: Device to check
2531 * @port_num: Port number to check
2533 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2534 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2535 * fabrics, devices should resolve routes to other hosts by contacting the
2536 * SA to query the proper route.
2538 * Return: true if the port should act as a client to the fabric Subnet
2539 * Administration interface. This does not imply that the SA service is
2542 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2544 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2548 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2550 * @device: Device to check
2551 * @port_num: Port number to check
2553 * InfiniBand multicast registration is more complex than normal IPv4 or
2554 * IPv6 multicast registration. Each Host Channel Adapter must register
2555 * with the Subnet Manager when it wishes to join a multicast group. It
2556 * should do so only once regardless of how many queue pairs it subscribes
2557 * to this group. And it should leave the group only after all queue pairs
2558 * attached to the group have been detached.
2560 * Return: true if the port must undertake the additional adminstrative
2561 * overhead of registering/unregistering with the SM and tracking of the
2562 * total number of queue pairs attached to the multicast group.
2564 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2566 return rdma_cap_ib_sa(device, port_num);
2570 * rdma_cap_af_ib - Check if the port of device has the capability
2571 * Native Infiniband Address.
2572 * @device: Device to check
2573 * @port_num: Port number to check
2575 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2576 * GID. RoCE uses a different mechanism, but still generates a GID via
2577 * a prescribed mechanism and port specific data.
2579 * Return: true if the port uses a GID address to identify devices on the
2582 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2584 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2588 * rdma_cap_eth_ah - Check if the port of device has the capability
2589 * Ethernet Address Handle.
2590 * @device: Device to check
2591 * @port_num: Port number to check
2593 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2594 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2595 * port. Normally, packet headers are generated by the sending host
2596 * adapter, but when sending connectionless datagrams, we must manually
2597 * inject the proper headers for the fabric we are communicating over.
2599 * Return: true if we are running as a RoCE port and must force the
2600 * addition of a Global Route Header built from our Ethernet Address
2601 * Handle into our header list for connectionless packets.
2603 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2605 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2609 * rdma_cap_opa_ah - Check if the port of device supports
2610 * OPA Address handles
2611 * @device: Device to check
2612 * @port_num: Port number to check
2614 * Return: true if we are running on an OPA device which supports
2615 * the extended OPA addressing.
2617 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2619 return (device->port_immutable[port_num].core_cap_flags &
2620 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2624 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2627 * @port_num: Port number
2629 * This MAD size includes the MAD headers and MAD payload. No other headers
2632 * Return the max MAD size required by the Port. Will return 0 if the port
2633 * does not support MADs
2635 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2637 return device->port_immutable[port_num].max_mad_size;
2641 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2642 * @device: Device to check
2643 * @port_num: Port number to check
2645 * RoCE GID table mechanism manages the various GIDs for a device.
2647 * NOTE: if allocating the port's GID table has failed, this call will still
2648 * return true, but any RoCE GID table API will fail.
2650 * Return: true if the port uses RoCE GID table mechanism in order to manage
2653 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2656 return rdma_protocol_roce(device, port_num) &&
2657 device->add_gid && device->del_gid;
2661 * Check if the device supports READ W/ INVALIDATE.
2663 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2666 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2667 * has support for it yet.
2669 return rdma_protocol_iwarp(dev, port_num);
2672 int ib_query_gid(struct ib_device *device,
2673 u8 port_num, int index, union ib_gid *gid,
2674 struct ib_gid_attr *attr);
2676 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2678 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2679 struct ifla_vf_info *info);
2680 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2681 struct ifla_vf_stats *stats);
2682 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2685 int ib_query_pkey(struct ib_device *device,
2686 u8 port_num, u16 index, u16 *pkey);
2688 int ib_modify_device(struct ib_device *device,
2689 int device_modify_mask,
2690 struct ib_device_modify *device_modify);
2692 int ib_modify_port(struct ib_device *device,
2693 u8 port_num, int port_modify_mask,
2694 struct ib_port_modify *port_modify);
2696 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2697 enum ib_gid_type gid_type, struct net_device *ndev,
2698 u8 *port_num, u16 *index);
2700 int ib_find_pkey(struct ib_device *device,
2701 u8 port_num, u16 pkey, u16 *index);
2705 * Create a memory registration for all memory in the system and place
2706 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2707 * ULPs to avoid the overhead of dynamic MRs.
2709 * This flag is generally considered unsafe and must only be used in
2710 * extremly trusted environments. Every use of it will log a warning
2711 * in the kernel log.
2713 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2716 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2717 const char *caller);
2718 #define ib_alloc_pd(device, flags) \
2719 __ib_alloc_pd((device), (flags), __func__)
2720 void ib_dealloc_pd(struct ib_pd *pd);
2723 * rdma_create_ah - Creates an address handle for the given address vector.
2724 * @pd: The protection domain associated with the address handle.
2725 * @ah_attr: The attributes of the address vector.
2727 * The address handle is used to reference a local or global destination
2728 * in all UD QP post sends.
2730 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2733 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2735 * @hdr: the L3 header to parse
2736 * @net_type: type of header to parse
2737 * @sgid: place to store source gid
2738 * @dgid: place to store destination gid
2740 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2741 enum rdma_network_type net_type,
2742 union ib_gid *sgid, union ib_gid *dgid);
2745 * ib_get_rdma_header_version - Get the header version
2746 * @hdr: the L3 header to parse
2748 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2751 * ib_init_ah_from_wc - Initializes address handle attributes from a
2753 * @device: Device on which the received message arrived.
2754 * @port_num: Port on which the received message arrived.
2755 * @wc: Work completion associated with the received message.
2756 * @grh: References the received global route header. This parameter is
2757 * ignored unless the work completion indicates that the GRH is valid.
2758 * @ah_attr: Returned attributes that can be used when creating an address
2759 * handle for replying to the message.
2761 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2762 const struct ib_wc *wc, const struct ib_grh *grh,
2763 struct rdma_ah_attr *ah_attr);
2766 * ib_create_ah_from_wc - Creates an address handle associated with the
2767 * sender of the specified work completion.
2768 * @pd: The protection domain associated with the address handle.
2769 * @wc: Work completion information associated with a received message.
2770 * @grh: References the received global route header. This parameter is
2771 * ignored unless the work completion indicates that the GRH is valid.
2772 * @port_num: The outbound port number to associate with the address.
2774 * The address handle is used to reference a local or global destination
2775 * in all UD QP post sends.
2777 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2778 const struct ib_grh *grh, u8 port_num);
2781 * rdma_modify_ah - Modifies the address vector associated with an address
2783 * @ah: The address handle to modify.
2784 * @ah_attr: The new address vector attributes to associate with the
2787 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2790 * rdma_query_ah - Queries the address vector associated with an address
2792 * @ah: The address handle to query.
2793 * @ah_attr: The address vector attributes associated with the address
2796 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2799 * rdma_destroy_ah - Destroys an address handle.
2800 * @ah: The address handle to destroy.
2802 int rdma_destroy_ah(struct ib_ah *ah);
2805 * ib_create_srq - Creates a SRQ associated with the specified protection
2807 * @pd: The protection domain associated with the SRQ.
2808 * @srq_init_attr: A list of initial attributes required to create the
2809 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2810 * the actual capabilities of the created SRQ.
2812 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2813 * requested size of the SRQ, and set to the actual values allocated
2814 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2815 * will always be at least as large as the requested values.
2817 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2818 struct ib_srq_init_attr *srq_init_attr);
2821 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2822 * @srq: The SRQ to modify.
2823 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2824 * the current values of selected SRQ attributes are returned.
2825 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2826 * are being modified.
2828 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2829 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2830 * the number of receives queued drops below the limit.
2832 int ib_modify_srq(struct ib_srq *srq,
2833 struct ib_srq_attr *srq_attr,
2834 enum ib_srq_attr_mask srq_attr_mask);
2837 * ib_query_srq - Returns the attribute list and current values for the
2839 * @srq: The SRQ to query.
2840 * @srq_attr: The attributes of the specified SRQ.
2842 int ib_query_srq(struct ib_srq *srq,
2843 struct ib_srq_attr *srq_attr);
2846 * ib_destroy_srq - Destroys the specified SRQ.
2847 * @srq: The SRQ to destroy.
2849 int ib_destroy_srq(struct ib_srq *srq);
2852 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2853 * @srq: The SRQ to post the work request on.
2854 * @recv_wr: A list of work requests to post on the receive queue.
2855 * @bad_recv_wr: On an immediate failure, this parameter will reference
2856 * the work request that failed to be posted on the QP.
2858 static inline int ib_post_srq_recv(struct ib_srq *srq,
2859 struct ib_recv_wr *recv_wr,
2860 struct ib_recv_wr **bad_recv_wr)
2862 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2866 * ib_create_qp - Creates a QP associated with the specified protection
2868 * @pd: The protection domain associated with the QP.
2869 * @qp_init_attr: A list of initial attributes required to create the
2870 * QP. If QP creation succeeds, then the attributes are updated to
2871 * the actual capabilities of the created QP.
2873 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2874 struct ib_qp_init_attr *qp_init_attr);
2877 * ib_modify_qp - Modifies the attributes for the specified QP and then
2878 * transitions the QP to the given state.
2879 * @qp: The QP to modify.
2880 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2881 * the current values of selected QP attributes are returned.
2882 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2883 * are being modified.
2885 int ib_modify_qp(struct ib_qp *qp,
2886 struct ib_qp_attr *qp_attr,
2890 * ib_query_qp - Returns the attribute list and current values for the
2892 * @qp: The QP to query.
2893 * @qp_attr: The attributes of the specified QP.
2894 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2895 * @qp_init_attr: Additional attributes of the selected QP.
2897 * The qp_attr_mask may be used to limit the query to gathering only the
2898 * selected attributes.
2900 int ib_query_qp(struct ib_qp *qp,
2901 struct ib_qp_attr *qp_attr,
2903 struct ib_qp_init_attr *qp_init_attr);
2906 * ib_destroy_qp - Destroys the specified QP.
2907 * @qp: The QP to destroy.
2909 int ib_destroy_qp(struct ib_qp *qp);
2912 * ib_open_qp - Obtain a reference to an existing sharable QP.
2913 * @xrcd - XRC domain
2914 * @qp_open_attr: Attributes identifying the QP to open.
2916 * Returns a reference to a sharable QP.
2918 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2919 struct ib_qp_open_attr *qp_open_attr);
2922 * ib_close_qp - Release an external reference to a QP.
2923 * @qp: The QP handle to release
2925 * The opened QP handle is released by the caller. The underlying
2926 * shared QP is not destroyed until all internal references are released.
2928 int ib_close_qp(struct ib_qp *qp);
2931 * ib_post_send - Posts a list of work requests to the send queue of
2933 * @qp: The QP to post the work request on.
2934 * @send_wr: A list of work requests to post on the send queue.
2935 * @bad_send_wr: On an immediate failure, this parameter will reference
2936 * the work request that failed to be posted on the QP.
2938 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2939 * error is returned, the QP state shall not be affected,
2940 * ib_post_send() will return an immediate error after queueing any
2941 * earlier work requests in the list.
2943 static inline int ib_post_send(struct ib_qp *qp,
2944 struct ib_send_wr *send_wr,
2945 struct ib_send_wr **bad_send_wr)
2947 return qp->device->post_send(qp, send_wr, bad_send_wr);
2951 * ib_post_recv - Posts a list of work requests to the receive queue of
2953 * @qp: The QP to post the work request on.
2954 * @recv_wr: A list of work requests to post on the receive queue.
2955 * @bad_recv_wr: On an immediate failure, this parameter will reference
2956 * the work request that failed to be posted on the QP.
2958 static inline int ib_post_recv(struct ib_qp *qp,
2959 struct ib_recv_wr *recv_wr,
2960 struct ib_recv_wr **bad_recv_wr)
2962 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2965 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2966 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2967 void ib_free_cq(struct ib_cq *cq);
2968 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2971 * ib_create_cq - Creates a CQ on the specified device.
2972 * @device: The device on which to create the CQ.
2973 * @comp_handler: A user-specified callback that is invoked when a
2974 * completion event occurs on the CQ.
2975 * @event_handler: A user-specified callback that is invoked when an
2976 * asynchronous event not associated with a completion occurs on the CQ.
2977 * @cq_context: Context associated with the CQ returned to the user via
2978 * the associated completion and event handlers.
2979 * @cq_attr: The attributes the CQ should be created upon.
2981 * Users can examine the cq structure to determine the actual CQ size.
2983 struct ib_cq *ib_create_cq(struct ib_device *device,
2984 ib_comp_handler comp_handler,
2985 void (*event_handler)(struct ib_event *, void *),
2987 const struct ib_cq_init_attr *cq_attr);
2990 * ib_resize_cq - Modifies the capacity of the CQ.
2991 * @cq: The CQ to resize.
2992 * @cqe: The minimum size of the CQ.
2994 * Users can examine the cq structure to determine the actual CQ size.
2996 int ib_resize_cq(struct ib_cq *cq, int cqe);
2999 * ib_modify_cq - Modifies moderation params of the CQ
3000 * @cq: The CQ to modify.
3001 * @cq_count: number of CQEs that will trigger an event
3002 * @cq_period: max period of time in usec before triggering an event
3005 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3008 * ib_destroy_cq - Destroys the specified CQ.
3009 * @cq: The CQ to destroy.
3011 int ib_destroy_cq(struct ib_cq *cq);
3014 * ib_poll_cq - poll a CQ for completion(s)
3015 * @cq:the CQ being polled
3016 * @num_entries:maximum number of completions to return
3017 * @wc:array of at least @num_entries &struct ib_wc where completions
3020 * Poll a CQ for (possibly multiple) completions. If the return value
3021 * is < 0, an error occurred. If the return value is >= 0, it is the
3022 * number of completions returned. If the return value is
3023 * non-negative and < num_entries, then the CQ was emptied.
3025 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3028 return cq->device->poll_cq(cq, num_entries, wc);
3032 * ib_peek_cq - Returns the number of unreaped completions currently
3033 * on the specified CQ.
3034 * @cq: The CQ to peek.
3035 * @wc_cnt: A minimum number of unreaped completions to check for.
3037 * If the number of unreaped completions is greater than or equal to wc_cnt,
3038 * this function returns wc_cnt, otherwise, it returns the actual number of
3039 * unreaped completions.
3041 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3044 * ib_req_notify_cq - Request completion notification on a CQ.
3045 * @cq: The CQ to generate an event for.
3047 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3048 * to request an event on the next solicited event or next work
3049 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3050 * may also be |ed in to request a hint about missed events, as
3054 * < 0 means an error occurred while requesting notification
3055 * == 0 means notification was requested successfully, and if
3056 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3057 * were missed and it is safe to wait for another event. In
3058 * this case is it guaranteed that any work completions added
3059 * to the CQ since the last CQ poll will trigger a completion
3060 * notification event.
3061 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3062 * in. It means that the consumer must poll the CQ again to
3063 * make sure it is empty to avoid missing an event because of a
3064 * race between requesting notification and an entry being
3065 * added to the CQ. This return value means it is possible
3066 * (but not guaranteed) that a work completion has been added
3067 * to the CQ since the last poll without triggering a
3068 * completion notification event.
3070 static inline int ib_req_notify_cq(struct ib_cq *cq,
3071 enum ib_cq_notify_flags flags)
3073 return cq->device->req_notify_cq(cq, flags);
3077 * ib_req_ncomp_notif - Request completion notification when there are
3078 * at least the specified number of unreaped completions on the CQ.
3079 * @cq: The CQ to generate an event for.
3080 * @wc_cnt: The number of unreaped completions that should be on the
3081 * CQ before an event is generated.
3083 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3085 return cq->device->req_ncomp_notif ?
3086 cq->device->req_ncomp_notif(cq, wc_cnt) :
3091 * ib_dma_mapping_error - check a DMA addr for error
3092 * @dev: The device for which the dma_addr was created
3093 * @dma_addr: The DMA address to check
3095 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3097 return dma_mapping_error(dev->dma_device, dma_addr);
3101 * ib_dma_map_single - Map a kernel virtual address to DMA address
3102 * @dev: The device for which the dma_addr is to be created
3103 * @cpu_addr: The kernel virtual address
3104 * @size: The size of the region in bytes
3105 * @direction: The direction of the DMA
3107 static inline u64 ib_dma_map_single(struct ib_device *dev,
3108 void *cpu_addr, size_t size,
3109 enum dma_data_direction direction)
3111 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3115 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3116 * @dev: The device for which the DMA address was created
3117 * @addr: The DMA address
3118 * @size: The size of the region in bytes
3119 * @direction: The direction of the DMA
3121 static inline void ib_dma_unmap_single(struct ib_device *dev,
3122 u64 addr, size_t size,
3123 enum dma_data_direction direction)
3125 dma_unmap_single(dev->dma_device, addr, size, direction);
3129 * ib_dma_map_page - Map a physical page to DMA address
3130 * @dev: The device for which the dma_addr is to be created
3131 * @page: The page to be mapped
3132 * @offset: The offset within the page
3133 * @size: The size of the region in bytes
3134 * @direction: The direction of the DMA
3136 static inline u64 ib_dma_map_page(struct ib_device *dev,
3138 unsigned long offset,
3140 enum dma_data_direction direction)
3142 return dma_map_page(dev->dma_device, page, offset, size, direction);
3146 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3147 * @dev: The device for which the DMA address was created
3148 * @addr: The DMA address
3149 * @size: The size of the region in bytes
3150 * @direction: The direction of the DMA
3152 static inline void ib_dma_unmap_page(struct ib_device *dev,
3153 u64 addr, size_t size,
3154 enum dma_data_direction direction)
3156 dma_unmap_page(dev->dma_device, addr, size, direction);
3160 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3161 * @dev: The device for which the DMA addresses are to be created
3162 * @sg: The array of scatter/gather entries
3163 * @nents: The number of scatter/gather entries
3164 * @direction: The direction of the DMA
3166 static inline int ib_dma_map_sg(struct ib_device *dev,
3167 struct scatterlist *sg, int nents,
3168 enum dma_data_direction direction)
3170 return dma_map_sg(dev->dma_device, sg, nents, direction);
3174 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3175 * @dev: The device for which the DMA addresses were created
3176 * @sg: The array of scatter/gather entries
3177 * @nents: The number of scatter/gather entries
3178 * @direction: The direction of the DMA
3180 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3181 struct scatterlist *sg, int nents,
3182 enum dma_data_direction direction)
3184 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3187 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3188 struct scatterlist *sg, int nents,
3189 enum dma_data_direction direction,
3190 unsigned long dma_attrs)
3192 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3196 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3197 struct scatterlist *sg, int nents,
3198 enum dma_data_direction direction,
3199 unsigned long dma_attrs)
3201 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3204 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3205 * @dev: The device for which the DMA addresses were created
3206 * @sg: The scatter/gather entry
3208 * Note: this function is obsolete. To do: change all occurrences of
3209 * ib_sg_dma_address() into sg_dma_address().
3211 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3212 struct scatterlist *sg)
3214 return sg_dma_address(sg);
3218 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3219 * @dev: The device for which the DMA addresses were created
3220 * @sg: The scatter/gather entry
3222 * Note: this function is obsolete. To do: change all occurrences of
3223 * ib_sg_dma_len() into sg_dma_len().
3225 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3226 struct scatterlist *sg)
3228 return sg_dma_len(sg);
3232 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3233 * @dev: The device for which the DMA address was created
3234 * @addr: The DMA address
3235 * @size: The size of the region in bytes
3236 * @dir: The direction of the DMA
3238 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3241 enum dma_data_direction dir)
3243 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3247 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3248 * @dev: The device for which the DMA address was created
3249 * @addr: The DMA address
3250 * @size: The size of the region in bytes
3251 * @dir: The direction of the DMA
3253 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3256 enum dma_data_direction dir)
3258 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3262 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3263 * @dev: The device for which the DMA address is requested
3264 * @size: The size of the region to allocate in bytes
3265 * @dma_handle: A pointer for returning the DMA address of the region
3266 * @flag: memory allocator flags
3268 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3270 dma_addr_t *dma_handle,
3273 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3277 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3278 * @dev: The device for which the DMA addresses were allocated
3279 * @size: The size of the region
3280 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3281 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3283 static inline void ib_dma_free_coherent(struct ib_device *dev,
3284 size_t size, void *cpu_addr,
3285 dma_addr_t dma_handle)
3287 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3291 * ib_dereg_mr - Deregisters a memory region and removes it from the
3292 * HCA translation table.
3293 * @mr: The memory region to deregister.
3295 * This function can fail, if the memory region has memory windows bound to it.
3297 int ib_dereg_mr(struct ib_mr *mr);
3299 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3300 enum ib_mr_type mr_type,
3304 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3306 * @mr - struct ib_mr pointer to be updated.
3307 * @newkey - new key to be used.
3309 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3311 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3312 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3316 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3317 * for calculating a new rkey for type 2 memory windows.
3318 * @rkey - the rkey to increment.
3320 static inline u32 ib_inc_rkey(u32 rkey)
3322 const u32 mask = 0x000000ff;
3323 return ((rkey + 1) & mask) | (rkey & ~mask);
3327 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3328 * @pd: The protection domain associated with the unmapped region.
3329 * @mr_access_flags: Specifies the memory access rights.
3330 * @fmr_attr: Attributes of the unmapped region.
3332 * A fast memory region must be mapped before it can be used as part of
3335 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3336 int mr_access_flags,
3337 struct ib_fmr_attr *fmr_attr);
3340 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3341 * @fmr: The fast memory region to associate with the pages.
3342 * @page_list: An array of physical pages to map to the fast memory region.
3343 * @list_len: The number of pages in page_list.
3344 * @iova: The I/O virtual address to use with the mapped region.
3346 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3347 u64 *page_list, int list_len,
3350 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3354 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3355 * @fmr_list: A linked list of fast memory regions to unmap.
3357 int ib_unmap_fmr(struct list_head *fmr_list);
3360 * ib_dealloc_fmr - Deallocates a fast memory region.
3361 * @fmr: The fast memory region to deallocate.
3363 int ib_dealloc_fmr(struct ib_fmr *fmr);
3366 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3367 * @qp: QP to attach to the multicast group. The QP must be type
3369 * @gid: Multicast group GID.
3370 * @lid: Multicast group LID in host byte order.
3372 * In order to send and receive multicast packets, subnet
3373 * administration must have created the multicast group and configured
3374 * the fabric appropriately. The port associated with the specified
3375 * QP must also be a member of the multicast group.
3377 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3380 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3381 * @qp: QP to detach from the multicast group.
3382 * @gid: Multicast group GID.
3383 * @lid: Multicast group LID in host byte order.
3385 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3388 * ib_alloc_xrcd - Allocates an XRC domain.
3389 * @device: The device on which to allocate the XRC domain.
3391 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3394 * ib_dealloc_xrcd - Deallocates an XRC domain.
3395 * @xrcd: The XRC domain to deallocate.
3397 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3399 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3400 struct ib_flow_attr *flow_attr, int domain);
3401 int ib_destroy_flow(struct ib_flow *flow_id);
3403 static inline int ib_check_mr_access(int flags)
3406 * Local write permission is required if remote write or
3407 * remote atomic permission is also requested.
3409 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3410 !(flags & IB_ACCESS_LOCAL_WRITE))
3417 * ib_check_mr_status: lightweight check of MR status.
3418 * This routine may provide status checks on a selected
3419 * ib_mr. first use is for signature status check.
3421 * @mr: A memory region.
3422 * @check_mask: Bitmask of which checks to perform from
3423 * ib_mr_status_check enumeration.
3424 * @mr_status: The container of relevant status checks.
3425 * failed checks will be indicated in the status bitmask
3426 * and the relevant info shall be in the error item.
3428 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3429 struct ib_mr_status *mr_status);
3431 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3432 u16 pkey, const union ib_gid *gid,
3433 const struct sockaddr *addr);
3434 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3435 struct ib_wq_init_attr *init_attr);
3436 int ib_destroy_wq(struct ib_wq *wq);
3437 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3439 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3440 struct ib_rwq_ind_table_init_attr*
3441 wq_ind_table_init_attr);
3442 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3444 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3445 unsigned int *sg_offset, unsigned int page_size);
3448 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3449 unsigned int *sg_offset, unsigned int page_size)
3453 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3459 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3460 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3462 void ib_drain_rq(struct ib_qp *qp);
3463 void ib_drain_sq(struct ib_qp *qp);
3464 void ib_drain_qp(struct ib_qp *qp);
3466 int ib_resolve_eth_dmac(struct ib_device *device,
3467 struct rdma_ah_attr *ah_attr);
3469 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3474 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3476 attr->dlid = (u16)dlid;
3479 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3484 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3489 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3494 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3497 attr->src_path_bits = src_path_bits;
3500 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3502 return attr->src_path_bits;
3505 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3507 attr->port_num = port_num;
3510 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3512 return attr->port_num;
3515 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3518 attr->static_rate = static_rate;
3521 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3523 return attr->static_rate;
3526 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3527 enum ib_ah_flags flag)
3529 attr->ah_flags = flag;
3532 static inline enum ib_ah_flags
3533 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3535 return attr->ah_flags;
3538 static inline const struct ib_global_route
3539 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3544 /*To retrieve and modify the grh */
3545 static inline struct ib_global_route
3546 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3551 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3553 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3555 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3558 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3561 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3563 grh->dgid.global.subnet_prefix = prefix;
3566 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3569 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3571 grh->dgid.global.interface_id = if_id;
3574 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3575 union ib_gid *dgid, u32 flow_label,
3576 u8 sgid_index, u8 hop_limit,
3579 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3581 attr->ah_flags = IB_AH_GRH;
3584 grh->flow_label = flow_label;
3585 grh->sgid_index = sgid_index;
3586 grh->hop_limit = hop_limit;
3587 grh->traffic_class = traffic_class;
3589 #endif /* IB_VERBS_H */