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 enum rdma_ah_attr_type {
844 RDMA_AH_ATTR_TYPE_IB,
845 RDMA_AH_ATTR_TYPE_ROCE,
846 RDMA_AH_ATTR_TYPE_OPA,
854 struct roce_ah_attr {
863 struct rdma_ah_attr {
864 struct ib_global_route grh;
869 enum rdma_ah_attr_type type;
871 struct ib_ah_attr ib;
872 struct roce_ah_attr roce;
873 struct opa_ah_attr opa;
881 IB_WC_LOC_EEC_OP_ERR,
886 IB_WC_LOC_ACCESS_ERR,
887 IB_WC_REM_INV_REQ_ERR,
888 IB_WC_REM_ACCESS_ERR,
891 IB_WC_RNR_RETRY_EXC_ERR,
892 IB_WC_LOC_RDD_VIOL_ERR,
893 IB_WC_REM_INV_RD_REQ_ERR,
896 IB_WC_INV_EEC_STATE_ERR,
898 IB_WC_RESP_TIMEOUT_ERR,
902 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
913 IB_WC_MASKED_COMP_SWAP,
914 IB_WC_MASKED_FETCH_ADD,
916 * Set value of IB_WC_RECV so consumers can test if a completion is a
917 * receive by testing (opcode & IB_WC_RECV).
920 IB_WC_RECV_RDMA_WITH_IMM
925 IB_WC_WITH_IMM = (1<<1),
926 IB_WC_WITH_INVALIDATE = (1<<2),
927 IB_WC_IP_CSUM_OK = (1<<3),
928 IB_WC_WITH_SMAC = (1<<4),
929 IB_WC_WITH_VLAN = (1<<5),
930 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
936 struct ib_cqe *wr_cqe;
938 enum ib_wc_status status;
939 enum ib_wc_opcode opcode;
953 u8 port_num; /* valid only for DR SMPs on switches */
959 enum ib_cq_notify_flags {
960 IB_CQ_SOLICITED = 1 << 0,
961 IB_CQ_NEXT_COMP = 1 << 1,
962 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
963 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
971 enum ib_srq_attr_mask {
972 IB_SRQ_MAX_WR = 1 << 0,
973 IB_SRQ_LIMIT = 1 << 1,
982 struct ib_srq_init_attr {
983 void (*event_handler)(struct ib_event *, void *);
985 struct ib_srq_attr attr;
986 enum ib_srq_type srq_type;
990 struct ib_xrcd *xrcd;
1001 u32 max_inline_data;
1004 * Maximum number of rdma_rw_ctx structures in flight at a time.
1005 * ib_create_qp() will calculate the right amount of neededed WRs
1006 * and MRs based on this.
1018 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1019 * here (and in that order) since the MAD layer uses them as
1020 * indices into a 2-entry table.
1029 IB_QPT_RAW_ETHERTYPE,
1030 IB_QPT_RAW_PACKET = 8,
1034 /* Reserve a range for qp types internal to the low level driver.
1035 * These qp types will not be visible at the IB core layer, so the
1036 * IB_QPT_MAX usages should not be affected in the core layer
1038 IB_QPT_RESERVED1 = 0x1000,
1050 enum ib_qp_create_flags {
1051 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1052 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1053 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1054 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1055 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1056 IB_QP_CREATE_NETIF_QP = 1 << 5,
1057 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1058 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
1059 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1060 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1061 /* reserve bits 26-31 for low level drivers' internal use */
1062 IB_QP_CREATE_RESERVED_START = 1 << 26,
1063 IB_QP_CREATE_RESERVED_END = 1 << 31,
1067 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1068 * callback to destroy the passed in QP.
1071 struct ib_qp_init_attr {
1072 void (*event_handler)(struct ib_event *, void *);
1074 struct ib_cq *send_cq;
1075 struct ib_cq *recv_cq;
1077 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1078 struct ib_qp_cap cap;
1079 enum ib_sig_type sq_sig_type;
1080 enum ib_qp_type qp_type;
1081 enum ib_qp_create_flags create_flags;
1084 * Only needed for special QP types, or when using the RW API.
1087 struct ib_rwq_ind_table *rwq_ind_tbl;
1090 struct ib_qp_open_attr {
1091 void (*event_handler)(struct ib_event *, void *);
1094 enum ib_qp_type qp_type;
1097 enum ib_rnr_timeout {
1098 IB_RNR_TIMER_655_36 = 0,
1099 IB_RNR_TIMER_000_01 = 1,
1100 IB_RNR_TIMER_000_02 = 2,
1101 IB_RNR_TIMER_000_03 = 3,
1102 IB_RNR_TIMER_000_04 = 4,
1103 IB_RNR_TIMER_000_06 = 5,
1104 IB_RNR_TIMER_000_08 = 6,
1105 IB_RNR_TIMER_000_12 = 7,
1106 IB_RNR_TIMER_000_16 = 8,
1107 IB_RNR_TIMER_000_24 = 9,
1108 IB_RNR_TIMER_000_32 = 10,
1109 IB_RNR_TIMER_000_48 = 11,
1110 IB_RNR_TIMER_000_64 = 12,
1111 IB_RNR_TIMER_000_96 = 13,
1112 IB_RNR_TIMER_001_28 = 14,
1113 IB_RNR_TIMER_001_92 = 15,
1114 IB_RNR_TIMER_002_56 = 16,
1115 IB_RNR_TIMER_003_84 = 17,
1116 IB_RNR_TIMER_005_12 = 18,
1117 IB_RNR_TIMER_007_68 = 19,
1118 IB_RNR_TIMER_010_24 = 20,
1119 IB_RNR_TIMER_015_36 = 21,
1120 IB_RNR_TIMER_020_48 = 22,
1121 IB_RNR_TIMER_030_72 = 23,
1122 IB_RNR_TIMER_040_96 = 24,
1123 IB_RNR_TIMER_061_44 = 25,
1124 IB_RNR_TIMER_081_92 = 26,
1125 IB_RNR_TIMER_122_88 = 27,
1126 IB_RNR_TIMER_163_84 = 28,
1127 IB_RNR_TIMER_245_76 = 29,
1128 IB_RNR_TIMER_327_68 = 30,
1129 IB_RNR_TIMER_491_52 = 31
1132 enum ib_qp_attr_mask {
1134 IB_QP_CUR_STATE = (1<<1),
1135 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1136 IB_QP_ACCESS_FLAGS = (1<<3),
1137 IB_QP_PKEY_INDEX = (1<<4),
1138 IB_QP_PORT = (1<<5),
1139 IB_QP_QKEY = (1<<6),
1141 IB_QP_PATH_MTU = (1<<8),
1142 IB_QP_TIMEOUT = (1<<9),
1143 IB_QP_RETRY_CNT = (1<<10),
1144 IB_QP_RNR_RETRY = (1<<11),
1145 IB_QP_RQ_PSN = (1<<12),
1146 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1147 IB_QP_ALT_PATH = (1<<14),
1148 IB_QP_MIN_RNR_TIMER = (1<<15),
1149 IB_QP_SQ_PSN = (1<<16),
1150 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1151 IB_QP_PATH_MIG_STATE = (1<<18),
1152 IB_QP_CAP = (1<<19),
1153 IB_QP_DEST_QPN = (1<<20),
1154 IB_QP_RESERVED1 = (1<<21),
1155 IB_QP_RESERVED2 = (1<<22),
1156 IB_QP_RESERVED3 = (1<<23),
1157 IB_QP_RESERVED4 = (1<<24),
1158 IB_QP_RATE_LIMIT = (1<<25),
1183 enum ib_qp_state qp_state;
1184 enum ib_qp_state cur_qp_state;
1185 enum ib_mtu path_mtu;
1186 enum ib_mig_state path_mig_state;
1191 int qp_access_flags;
1192 struct ib_qp_cap cap;
1193 struct rdma_ah_attr ah_attr;
1194 struct rdma_ah_attr alt_ah_attr;
1197 u8 en_sqd_async_notify;
1200 u8 max_dest_rd_atomic;
1213 IB_WR_RDMA_WRITE_WITH_IMM,
1215 IB_WR_SEND_WITH_IMM,
1217 IB_WR_ATOMIC_CMP_AND_SWP,
1218 IB_WR_ATOMIC_FETCH_AND_ADD,
1220 IB_WR_SEND_WITH_INV,
1221 IB_WR_RDMA_READ_WITH_INV,
1224 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1225 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1227 /* reserve values for low level drivers' internal use.
1228 * These values will not be used at all in the ib core layer.
1230 IB_WR_RESERVED1 = 0xf0,
1242 enum ib_send_flags {
1244 IB_SEND_SIGNALED = (1<<1),
1245 IB_SEND_SOLICITED = (1<<2),
1246 IB_SEND_INLINE = (1<<3),
1247 IB_SEND_IP_CSUM = (1<<4),
1249 /* reserve bits 26-31 for low level drivers' internal use */
1250 IB_SEND_RESERVED_START = (1 << 26),
1251 IB_SEND_RESERVED_END = (1 << 31),
1261 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1265 struct ib_send_wr *next;
1268 struct ib_cqe *wr_cqe;
1270 struct ib_sge *sg_list;
1272 enum ib_wr_opcode opcode;
1276 u32 invalidate_rkey;
1281 struct ib_send_wr wr;
1286 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1288 return container_of(wr, struct ib_rdma_wr, wr);
1291 struct ib_atomic_wr {
1292 struct ib_send_wr wr;
1296 u64 compare_add_mask;
1301 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1303 return container_of(wr, struct ib_atomic_wr, wr);
1307 struct ib_send_wr wr;
1314 u16 pkey_index; /* valid for GSI only */
1315 u8 port_num; /* valid for DR SMPs on switch only */
1318 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1320 return container_of(wr, struct ib_ud_wr, wr);
1324 struct ib_send_wr wr;
1330 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1332 return container_of(wr, struct ib_reg_wr, wr);
1335 struct ib_sig_handover_wr {
1336 struct ib_send_wr wr;
1337 struct ib_sig_attrs *sig_attrs;
1338 struct ib_mr *sig_mr;
1340 struct ib_sge *prot;
1343 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1345 return container_of(wr, struct ib_sig_handover_wr, wr);
1349 struct ib_recv_wr *next;
1352 struct ib_cqe *wr_cqe;
1354 struct ib_sge *sg_list;
1358 enum ib_access_flags {
1359 IB_ACCESS_LOCAL_WRITE = 1,
1360 IB_ACCESS_REMOTE_WRITE = (1<<1),
1361 IB_ACCESS_REMOTE_READ = (1<<2),
1362 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1363 IB_ACCESS_MW_BIND = (1<<4),
1364 IB_ZERO_BASED = (1<<5),
1365 IB_ACCESS_ON_DEMAND = (1<<6),
1366 IB_ACCESS_HUGETLB = (1<<7),
1370 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1371 * are hidden here instead of a uapi header!
1373 enum ib_mr_rereg_flags {
1374 IB_MR_REREG_TRANS = 1,
1375 IB_MR_REREG_PD = (1<<1),
1376 IB_MR_REREG_ACCESS = (1<<2),
1377 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1380 struct ib_fmr_attr {
1388 enum rdma_remove_reason {
1389 /* Userspace requested uobject deletion. Call could fail */
1390 RDMA_REMOVE_DESTROY,
1391 /* Context deletion. This call should delete the actual object itself */
1393 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1394 RDMA_REMOVE_DRIVER_REMOVE,
1395 /* Context is being cleaned-up, but commit was just completed */
1396 RDMA_REMOVE_DURING_CLEANUP,
1399 struct ib_rdmacg_object {
1400 #ifdef CONFIG_CGROUP_RDMA
1401 struct rdma_cgroup *cg; /* owner rdma cgroup */
1405 struct ib_ucontext {
1406 struct ib_device *device;
1407 struct ib_uverbs_file *ufile;
1410 /* locking the uobjects_list */
1411 struct mutex uobjects_lock;
1412 struct list_head uobjects;
1413 /* protects cleanup process from other actions */
1414 struct rw_semaphore cleanup_rwsem;
1415 enum rdma_remove_reason cleanup_reason;
1418 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1419 struct rb_root umem_tree;
1421 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1422 * mmu notifiers registration.
1424 struct rw_semaphore umem_rwsem;
1425 void (*invalidate_range)(struct ib_umem *umem,
1426 unsigned long start, unsigned long end);
1428 struct mmu_notifier mn;
1429 atomic_t notifier_count;
1430 /* A list of umems that don't have private mmu notifier counters yet. */
1431 struct list_head no_private_counters;
1435 struct ib_rdmacg_object cg_obj;
1439 u64 user_handle; /* handle given to us by userspace */
1440 struct ib_ucontext *context; /* associated user context */
1441 void *object; /* containing object */
1442 struct list_head list; /* link to context's list */
1443 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1444 int id; /* index into kernel idr */
1446 atomic_t usecnt; /* protects exclusive access */
1447 struct rcu_head rcu; /* kfree_rcu() overhead */
1449 const struct uverbs_obj_type *type;
1452 struct ib_uobject_file {
1453 struct ib_uobject uobj;
1454 /* ufile contains the lock between context release and file close */
1455 struct ib_uverbs_file *ufile;
1459 const void __user *inbuf;
1460 void __user *outbuf;
1468 struct ib_device *device;
1469 struct ib_uobject *uobject;
1470 atomic_t usecnt; /* count all resources */
1472 u32 unsafe_global_rkey;
1475 * Implementation details of the RDMA core, don't use in drivers:
1477 struct ib_mr *__internal_mr;
1481 struct ib_device *device;
1482 atomic_t usecnt; /* count all exposed resources */
1483 struct inode *inode;
1485 struct mutex tgt_qp_mutex;
1486 struct list_head tgt_qp_list;
1490 struct ib_device *device;
1492 struct ib_uobject *uobject;
1493 enum rdma_ah_attr_type type;
1496 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1498 enum ib_poll_context {
1499 IB_POLL_DIRECT, /* caller context, no hw completions */
1500 IB_POLL_SOFTIRQ, /* poll from softirq context */
1501 IB_POLL_WORKQUEUE, /* poll from workqueue */
1505 struct ib_device *device;
1506 struct ib_uobject *uobject;
1507 ib_comp_handler comp_handler;
1508 void (*event_handler)(struct ib_event *, void *);
1511 atomic_t usecnt; /* count number of work queues */
1512 enum ib_poll_context poll_ctx;
1515 struct irq_poll iop;
1516 struct work_struct work;
1521 struct ib_device *device;
1523 struct ib_uobject *uobject;
1524 void (*event_handler)(struct ib_event *, void *);
1526 enum ib_srq_type srq_type;
1531 struct ib_xrcd *xrcd;
1538 enum ib_raw_packet_caps {
1539 /* Strip cvlan from incoming packet and report it in the matching work
1540 * completion is supported.
1542 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1543 /* Scatter FCS field of an incoming packet to host memory is supported.
1545 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1546 /* Checksum offloads are supported (for both send and receive). */
1547 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1561 struct ib_device *device;
1562 struct ib_uobject *uobject;
1564 void (*event_handler)(struct ib_event *, void *);
1568 enum ib_wq_state state;
1569 enum ib_wq_type wq_type;
1574 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1575 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1578 struct ib_wq_init_attr {
1580 enum ib_wq_type wq_type;
1584 void (*event_handler)(struct ib_event *, void *);
1585 u32 create_flags; /* Use enum ib_wq_flags */
1588 enum ib_wq_attr_mask {
1589 IB_WQ_STATE = 1 << 0,
1590 IB_WQ_CUR_STATE = 1 << 1,
1591 IB_WQ_FLAGS = 1 << 2,
1595 enum ib_wq_state wq_state;
1596 enum ib_wq_state curr_wq_state;
1597 u32 flags; /* Use enum ib_wq_flags */
1598 u32 flags_mask; /* Use enum ib_wq_flags */
1601 struct ib_rwq_ind_table {
1602 struct ib_device *device;
1603 struct ib_uobject *uobject;
1606 u32 log_ind_tbl_size;
1607 struct ib_wq **ind_tbl;
1610 struct ib_rwq_ind_table_init_attr {
1611 u32 log_ind_tbl_size;
1612 /* Each entry is a pointer to Receive Work Queue */
1613 struct ib_wq **ind_tbl;
1617 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1618 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1621 struct ib_device *device;
1623 struct ib_cq *send_cq;
1624 struct ib_cq *recv_cq;
1627 struct list_head rdma_mrs;
1628 struct list_head sig_mrs;
1630 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1631 struct list_head xrcd_list;
1633 /* count times opened, mcast attaches, flow attaches */
1635 struct list_head open_list;
1636 struct ib_qp *real_qp;
1637 struct ib_uobject *uobject;
1638 void (*event_handler)(struct ib_event *, void *);
1643 enum ib_qp_type qp_type;
1644 struct ib_rwq_ind_table *rwq_ind_tbl;
1648 struct ib_device *device;
1654 unsigned int page_size;
1657 struct ib_uobject *uobject; /* user */
1658 struct list_head qp_entry; /* FR */
1663 struct ib_device *device;
1665 struct ib_uobject *uobject;
1667 enum ib_mw_type type;
1671 struct ib_device *device;
1673 struct list_head list;
1678 /* Supported steering options */
1679 enum ib_flow_attr_type {
1680 /* steering according to rule specifications */
1681 IB_FLOW_ATTR_NORMAL = 0x0,
1682 /* default unicast and multicast rule -
1683 * receive all Eth traffic which isn't steered to any QP
1685 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1686 /* default multicast rule -
1687 * receive all Eth multicast traffic which isn't steered to any QP
1689 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1690 /* sniffer rule - receive all port traffic */
1691 IB_FLOW_ATTR_SNIFFER = 0x3
1694 /* Supported steering header types */
1695 enum ib_flow_spec_type {
1697 IB_FLOW_SPEC_ETH = 0x20,
1698 IB_FLOW_SPEC_IB = 0x22,
1700 IB_FLOW_SPEC_IPV4 = 0x30,
1701 IB_FLOW_SPEC_IPV6 = 0x31,
1703 IB_FLOW_SPEC_TCP = 0x40,
1704 IB_FLOW_SPEC_UDP = 0x41,
1705 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1706 IB_FLOW_SPEC_INNER = 0x100,
1708 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1709 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1711 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1712 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1714 /* Flow steering rule priority is set according to it's domain.
1715 * Lower domain value means higher priority.
1717 enum ib_flow_domain {
1718 IB_FLOW_DOMAIN_USER,
1719 IB_FLOW_DOMAIN_ETHTOOL,
1722 IB_FLOW_DOMAIN_NUM /* Must be last */
1725 enum ib_flow_flags {
1726 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1727 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1730 struct ib_flow_eth_filter {
1739 struct ib_flow_spec_eth {
1742 struct ib_flow_eth_filter val;
1743 struct ib_flow_eth_filter mask;
1746 struct ib_flow_ib_filter {
1753 struct ib_flow_spec_ib {
1756 struct ib_flow_ib_filter val;
1757 struct ib_flow_ib_filter mask;
1760 /* IPv4 header flags */
1761 enum ib_ipv4_flags {
1762 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1763 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1764 last have this flag set */
1767 struct ib_flow_ipv4_filter {
1778 struct ib_flow_spec_ipv4 {
1781 struct ib_flow_ipv4_filter val;
1782 struct ib_flow_ipv4_filter mask;
1785 struct ib_flow_ipv6_filter {
1796 struct ib_flow_spec_ipv6 {
1799 struct ib_flow_ipv6_filter val;
1800 struct ib_flow_ipv6_filter mask;
1803 struct ib_flow_tcp_udp_filter {
1810 struct ib_flow_spec_tcp_udp {
1813 struct ib_flow_tcp_udp_filter val;
1814 struct ib_flow_tcp_udp_filter mask;
1817 struct ib_flow_tunnel_filter {
1822 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1823 * the tunnel_id from val has the vni value
1825 struct ib_flow_spec_tunnel {
1828 struct ib_flow_tunnel_filter val;
1829 struct ib_flow_tunnel_filter mask;
1832 struct ib_flow_spec_action_tag {
1833 enum ib_flow_spec_type type;
1838 struct ib_flow_spec_action_drop {
1839 enum ib_flow_spec_type type;
1843 union ib_flow_spec {
1848 struct ib_flow_spec_eth eth;
1849 struct ib_flow_spec_ib ib;
1850 struct ib_flow_spec_ipv4 ipv4;
1851 struct ib_flow_spec_tcp_udp tcp_udp;
1852 struct ib_flow_spec_ipv6 ipv6;
1853 struct ib_flow_spec_tunnel tunnel;
1854 struct ib_flow_spec_action_tag flow_tag;
1855 struct ib_flow_spec_action_drop drop;
1858 struct ib_flow_attr {
1859 enum ib_flow_attr_type type;
1865 /* Following are the optional layers according to user request
1866 * struct ib_flow_spec_xxx
1867 * struct ib_flow_spec_yyy
1873 struct ib_uobject *uobject;
1879 enum ib_process_mad_flags {
1880 IB_MAD_IGNORE_MKEY = 1,
1881 IB_MAD_IGNORE_BKEY = 2,
1882 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1885 enum ib_mad_result {
1886 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1887 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1888 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1889 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1892 #define IB_DEVICE_NAME_MAX 64
1894 struct ib_port_cache {
1895 struct ib_pkey_cache *pkey;
1896 struct ib_gid_table *gid;
1898 enum ib_port_state port_state;
1903 struct ib_event_handler event_handler;
1904 struct ib_port_cache *ports;
1909 struct ib_port_immutable {
1916 /* rdma netdev type - specifies protocol type */
1917 enum rdma_netdev_t {
1918 RDMA_NETDEV_OPA_VNIC,
1923 * struct rdma_netdev - rdma netdev
1924 * For cases where netstack interfacing is required.
1926 struct rdma_netdev {
1928 struct ib_device *hca;
1931 /* control functions */
1932 void (*set_id)(struct net_device *netdev, int id);
1934 int (*send)(struct net_device *dev, struct sk_buff *skb,
1935 struct ib_ah *address, u32 dqpn);
1937 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
1938 union ib_gid *gid, u16 mlid,
1939 int set_qkey, u32 qkey);
1940 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
1941 union ib_gid *gid, u16 mlid);
1945 /* Do not access @dma_device directly from ULP nor from HW drivers. */
1946 struct device *dma_device;
1948 char name[IB_DEVICE_NAME_MAX];
1950 struct list_head event_handler_list;
1951 spinlock_t event_handler_lock;
1953 spinlock_t client_data_lock;
1954 struct list_head core_list;
1955 /* Access to the client_data_list is protected by the client_data_lock
1956 * spinlock and the lists_rwsem read-write semaphore */
1957 struct list_head client_data_list;
1959 struct ib_cache cache;
1961 * port_immutable is indexed by port number
1963 struct ib_port_immutable *port_immutable;
1965 int num_comp_vectors;
1967 struct iw_cm_verbs *iwcm;
1970 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1971 * driver initialized data. The struct is kfree()'ed by the sysfs
1972 * core when the device is removed. A lifespan of -1 in the return
1973 * struct tells the core to set a default lifespan.
1975 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1978 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1979 * @index - The index in the value array we wish to have updated, or
1980 * num_counters if we want all stats updated
1982 * < 0 - Error, no counters updated
1983 * index - Updated the single counter pointed to by index
1984 * num_counters - Updated all counters (will reset the timestamp
1985 * and prevent further calls for lifespan milliseconds)
1986 * Drivers are allowed to update all counters in leiu of just the
1987 * one given in index at their option
1989 int (*get_hw_stats)(struct ib_device *device,
1990 struct rdma_hw_stats *stats,
1991 u8 port, int index);
1992 int (*query_device)(struct ib_device *device,
1993 struct ib_device_attr *device_attr,
1994 struct ib_udata *udata);
1995 int (*query_port)(struct ib_device *device,
1997 struct ib_port_attr *port_attr);
1998 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2000 /* When calling get_netdev, the HW vendor's driver should return the
2001 * net device of device @device at port @port_num or NULL if such
2002 * a net device doesn't exist. The vendor driver should call dev_hold
2003 * on this net device. The HW vendor's device driver must guarantee
2004 * that this function returns NULL before the net device reaches
2005 * NETDEV_UNREGISTER_FINAL state.
2007 struct net_device *(*get_netdev)(struct ib_device *device,
2009 int (*query_gid)(struct ib_device *device,
2010 u8 port_num, int index,
2012 /* When calling add_gid, the HW vendor's driver should
2013 * add the gid of device @device at gid index @index of
2014 * port @port_num to be @gid. Meta-info of that gid (for example,
2015 * the network device related to this gid is available
2016 * at @attr. @context allows the HW vendor driver to store extra
2017 * information together with a GID entry. The HW vendor may allocate
2018 * memory to contain this information and store it in @context when a
2019 * new GID entry is written to. Params are consistent until the next
2020 * call of add_gid or delete_gid. The function should return 0 on
2021 * success or error otherwise. The function could be called
2022 * concurrently for different ports. This function is only called
2023 * when roce_gid_table is used.
2025 int (*add_gid)(struct ib_device *device,
2028 const union ib_gid *gid,
2029 const struct ib_gid_attr *attr,
2031 /* When calling del_gid, the HW vendor's driver should delete the
2032 * gid of device @device at gid index @index of port @port_num.
2033 * Upon the deletion of a GID entry, the HW vendor must free any
2034 * allocated memory. The caller will clear @context afterwards.
2035 * This function is only called when roce_gid_table is used.
2037 int (*del_gid)(struct ib_device *device,
2041 int (*query_pkey)(struct ib_device *device,
2042 u8 port_num, u16 index, u16 *pkey);
2043 int (*modify_device)(struct ib_device *device,
2044 int device_modify_mask,
2045 struct ib_device_modify *device_modify);
2046 int (*modify_port)(struct ib_device *device,
2047 u8 port_num, int port_modify_mask,
2048 struct ib_port_modify *port_modify);
2049 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2050 struct ib_udata *udata);
2051 int (*dealloc_ucontext)(struct ib_ucontext *context);
2052 int (*mmap)(struct ib_ucontext *context,
2053 struct vm_area_struct *vma);
2054 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2055 struct ib_ucontext *context,
2056 struct ib_udata *udata);
2057 int (*dealloc_pd)(struct ib_pd *pd);
2058 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2059 struct rdma_ah_attr *ah_attr,
2060 struct ib_udata *udata);
2061 int (*modify_ah)(struct ib_ah *ah,
2062 struct rdma_ah_attr *ah_attr);
2063 int (*query_ah)(struct ib_ah *ah,
2064 struct rdma_ah_attr *ah_attr);
2065 int (*destroy_ah)(struct ib_ah *ah);
2066 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2067 struct ib_srq_init_attr *srq_init_attr,
2068 struct ib_udata *udata);
2069 int (*modify_srq)(struct ib_srq *srq,
2070 struct ib_srq_attr *srq_attr,
2071 enum ib_srq_attr_mask srq_attr_mask,
2072 struct ib_udata *udata);
2073 int (*query_srq)(struct ib_srq *srq,
2074 struct ib_srq_attr *srq_attr);
2075 int (*destroy_srq)(struct ib_srq *srq);
2076 int (*post_srq_recv)(struct ib_srq *srq,
2077 struct ib_recv_wr *recv_wr,
2078 struct ib_recv_wr **bad_recv_wr);
2079 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2080 struct ib_qp_init_attr *qp_init_attr,
2081 struct ib_udata *udata);
2082 int (*modify_qp)(struct ib_qp *qp,
2083 struct ib_qp_attr *qp_attr,
2085 struct ib_udata *udata);
2086 int (*query_qp)(struct ib_qp *qp,
2087 struct ib_qp_attr *qp_attr,
2089 struct ib_qp_init_attr *qp_init_attr);
2090 int (*destroy_qp)(struct ib_qp *qp);
2091 int (*post_send)(struct ib_qp *qp,
2092 struct ib_send_wr *send_wr,
2093 struct ib_send_wr **bad_send_wr);
2094 int (*post_recv)(struct ib_qp *qp,
2095 struct ib_recv_wr *recv_wr,
2096 struct ib_recv_wr **bad_recv_wr);
2097 struct ib_cq * (*create_cq)(struct ib_device *device,
2098 const struct ib_cq_init_attr *attr,
2099 struct ib_ucontext *context,
2100 struct ib_udata *udata);
2101 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2103 int (*destroy_cq)(struct ib_cq *cq);
2104 int (*resize_cq)(struct ib_cq *cq, int cqe,
2105 struct ib_udata *udata);
2106 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2108 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2109 int (*req_notify_cq)(struct ib_cq *cq,
2110 enum ib_cq_notify_flags flags);
2111 int (*req_ncomp_notif)(struct ib_cq *cq,
2113 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2114 int mr_access_flags);
2115 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2116 u64 start, u64 length,
2118 int mr_access_flags,
2119 struct ib_udata *udata);
2120 int (*rereg_user_mr)(struct ib_mr *mr,
2122 u64 start, u64 length,
2124 int mr_access_flags,
2126 struct ib_udata *udata);
2127 int (*dereg_mr)(struct ib_mr *mr);
2128 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2129 enum ib_mr_type mr_type,
2131 int (*map_mr_sg)(struct ib_mr *mr,
2132 struct scatterlist *sg,
2134 unsigned int *sg_offset);
2135 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2136 enum ib_mw_type type,
2137 struct ib_udata *udata);
2138 int (*dealloc_mw)(struct ib_mw *mw);
2139 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2140 int mr_access_flags,
2141 struct ib_fmr_attr *fmr_attr);
2142 int (*map_phys_fmr)(struct ib_fmr *fmr,
2143 u64 *page_list, int list_len,
2145 int (*unmap_fmr)(struct list_head *fmr_list);
2146 int (*dealloc_fmr)(struct ib_fmr *fmr);
2147 int (*attach_mcast)(struct ib_qp *qp,
2150 int (*detach_mcast)(struct ib_qp *qp,
2153 int (*process_mad)(struct ib_device *device,
2154 int process_mad_flags,
2156 const struct ib_wc *in_wc,
2157 const struct ib_grh *in_grh,
2158 const struct ib_mad_hdr *in_mad,
2160 struct ib_mad_hdr *out_mad,
2161 size_t *out_mad_size,
2162 u16 *out_mad_pkey_index);
2163 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2164 struct ib_ucontext *ucontext,
2165 struct ib_udata *udata);
2166 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2167 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2171 int (*destroy_flow)(struct ib_flow *flow_id);
2172 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2173 struct ib_mr_status *mr_status);
2174 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2175 void (*drain_rq)(struct ib_qp *qp);
2176 void (*drain_sq)(struct ib_qp *qp);
2177 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2179 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2180 struct ifla_vf_info *ivf);
2181 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2182 struct ifla_vf_stats *stats);
2183 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2185 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2186 struct ib_wq_init_attr *init_attr,
2187 struct ib_udata *udata);
2188 int (*destroy_wq)(struct ib_wq *wq);
2189 int (*modify_wq)(struct ib_wq *wq,
2190 struct ib_wq_attr *attr,
2192 struct ib_udata *udata);
2193 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2194 struct ib_rwq_ind_table_init_attr *init_attr,
2195 struct ib_udata *udata);
2196 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2198 * rdma netdev operations
2200 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2201 * doesn't support the specified rdma netdev type.
2203 struct net_device *(*alloc_rdma_netdev)(
2204 struct ib_device *device,
2206 enum rdma_netdev_t type,
2208 unsigned char name_assign_type,
2209 void (*setup)(struct net_device *));
2210 void (*free_rdma_netdev)(struct net_device *netdev);
2212 struct module *owner;
2214 struct kobject *ports_parent;
2215 struct list_head port_list;
2218 IB_DEV_UNINITIALIZED,
2224 u64 uverbs_cmd_mask;
2225 u64 uverbs_ex_cmd_mask;
2227 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2233 struct ib_device_attr attrs;
2234 struct attribute_group *hw_stats_ag;
2235 struct rdma_hw_stats *hw_stats;
2237 #ifdef CONFIG_CGROUP_RDMA
2238 struct rdmacg_device cg_device;
2242 * The following mandatory functions are used only at device
2243 * registration. Keep functions such as these at the end of this
2244 * structure to avoid cache line misses when accessing struct ib_device
2247 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2248 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2253 void (*add) (struct ib_device *);
2254 void (*remove)(struct ib_device *, void *client_data);
2256 /* Returns the net_dev belonging to this ib_client and matching the
2258 * @dev: An RDMA device that the net_dev use for communication.
2259 * @port: A physical port number on the RDMA device.
2260 * @pkey: P_Key that the net_dev uses if applicable.
2261 * @gid: A GID that the net_dev uses to communicate.
2262 * @addr: An IP address the net_dev is configured with.
2263 * @client_data: The device's client data set by ib_set_client_data().
2265 * An ib_client that implements a net_dev on top of RDMA devices
2266 * (such as IP over IB) should implement this callback, allowing the
2267 * rdma_cm module to find the right net_dev for a given request.
2269 * The caller is responsible for calling dev_put on the returned
2271 struct net_device *(*get_net_dev_by_params)(
2272 struct ib_device *dev,
2275 const union ib_gid *gid,
2276 const struct sockaddr *addr,
2278 struct list_head list;
2281 struct ib_device *ib_alloc_device(size_t size);
2282 void ib_dealloc_device(struct ib_device *device);
2284 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2286 int ib_register_device(struct ib_device *device,
2287 int (*port_callback)(struct ib_device *,
2288 u8, struct kobject *));
2289 void ib_unregister_device(struct ib_device *device);
2291 int ib_register_client (struct ib_client *client);
2292 void ib_unregister_client(struct ib_client *client);
2294 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2295 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2298 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2300 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2303 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2305 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2308 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2312 const void __user *p = udata->inbuf + offset;
2316 if (len > USHRT_MAX)
2319 buf = memdup_user(p, len);
2323 ret = !memchr_inv(buf, 0, len);
2329 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2330 * contains all required attributes and no attributes not allowed for
2331 * the given QP state transition.
2332 * @cur_state: Current QP state
2333 * @next_state: Next QP state
2335 * @mask: Mask of supplied QP attributes
2336 * @ll : link layer of port
2338 * This function is a helper function that a low-level driver's
2339 * modify_qp method can use to validate the consumer's input. It
2340 * checks that cur_state and next_state are valid QP states, that a
2341 * transition from cur_state to next_state is allowed by the IB spec,
2342 * and that the attribute mask supplied is allowed for the transition.
2344 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2345 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2346 enum rdma_link_layer ll);
2348 int ib_register_event_handler (struct ib_event_handler *event_handler);
2349 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2350 void ib_dispatch_event(struct ib_event *event);
2352 int ib_query_port(struct ib_device *device,
2353 u8 port_num, struct ib_port_attr *port_attr);
2355 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2359 * rdma_cap_ib_switch - Check if the device is IB switch
2360 * @device: Device to check
2362 * Device driver is responsible for setting is_switch bit on
2363 * in ib_device structure at init time.
2365 * Return: true if the device is IB switch.
2367 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2369 return device->is_switch;
2373 * rdma_start_port - Return the first valid port number for the device
2376 * @device: Device to be checked
2378 * Return start port number
2380 static inline u8 rdma_start_port(const struct ib_device *device)
2382 return rdma_cap_ib_switch(device) ? 0 : 1;
2386 * rdma_end_port - Return the last valid port number for the device
2389 * @device: Device to be checked
2391 * Return last port number
2393 static inline u8 rdma_end_port(const struct ib_device *device)
2395 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2398 static inline int rdma_is_port_valid(const struct ib_device *device,
2401 return (port >= rdma_start_port(device) &&
2402 port <= rdma_end_port(device));
2405 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2407 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2410 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2412 return device->port_immutable[port_num].core_cap_flags &
2413 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2416 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2418 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2421 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2423 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2426 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2428 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2431 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2433 return rdma_protocol_ib(device, port_num) ||
2434 rdma_protocol_roce(device, port_num);
2437 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2439 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2442 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2444 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2448 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2449 * Management Datagrams.
2450 * @device: Device to check
2451 * @port_num: Port number to check
2453 * Management Datagrams (MAD) are a required part of the InfiniBand
2454 * specification and are supported on all InfiniBand devices. A slightly
2455 * extended version are also supported on OPA interfaces.
2457 * Return: true if the port supports sending/receiving of MAD packets.
2459 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2461 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2465 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2466 * Management Datagrams.
2467 * @device: Device to check
2468 * @port_num: Port number to check
2470 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2471 * datagrams with their own versions. These OPA MADs share many but not all of
2472 * the characteristics of InfiniBand MADs.
2474 * OPA MADs differ in the following ways:
2476 * 1) MADs are variable size up to 2K
2477 * IBTA defined MADs remain fixed at 256 bytes
2478 * 2) OPA SMPs must carry valid PKeys
2479 * 3) OPA SMP packets are a different format
2481 * Return: true if the port supports OPA MAD packet formats.
2483 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2485 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2486 == RDMA_CORE_CAP_OPA_MAD;
2490 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2491 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2492 * @device: Device to check
2493 * @port_num: Port number to check
2495 * Each InfiniBand node is required to provide a Subnet Management Agent
2496 * that the subnet manager can access. Prior to the fabric being fully
2497 * configured by the subnet manager, the SMA is accessed via a well known
2498 * interface called the Subnet Management Interface (SMI). This interface
2499 * uses directed route packets to communicate with the SM to get around the
2500 * chicken and egg problem of the SM needing to know what's on the fabric
2501 * in order to configure the fabric, and needing to configure the fabric in
2502 * order to send packets to the devices on the fabric. These directed
2503 * route packets do not need the fabric fully configured in order to reach
2504 * their destination. The SMI is the only method allowed to send
2505 * directed route packets on an InfiniBand fabric.
2507 * Return: true if the port provides an SMI.
2509 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2511 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2515 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2516 * Communication Manager.
2517 * @device: Device to check
2518 * @port_num: Port number to check
2520 * The InfiniBand Communication Manager is one of many pre-defined General
2521 * Service Agents (GSA) that are accessed via the General Service
2522 * Interface (GSI). It's role is to facilitate establishment of connections
2523 * between nodes as well as other management related tasks for established
2526 * Return: true if the port supports an IB CM (this does not guarantee that
2527 * a CM is actually running however).
2529 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2531 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2535 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2536 * Communication Manager.
2537 * @device: Device to check
2538 * @port_num: Port number to check
2540 * Similar to above, but specific to iWARP connections which have a different
2541 * managment protocol than InfiniBand.
2543 * Return: true if the port supports an iWARP CM (this does not guarantee that
2544 * a CM is actually running however).
2546 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2548 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2552 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2553 * Subnet Administration.
2554 * @device: Device to check
2555 * @port_num: Port number to check
2557 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2558 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2559 * fabrics, devices should resolve routes to other hosts by contacting the
2560 * SA to query the proper route.
2562 * Return: true if the port should act as a client to the fabric Subnet
2563 * Administration interface. This does not imply that the SA service is
2566 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2568 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2572 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2574 * @device: Device to check
2575 * @port_num: Port number to check
2577 * InfiniBand multicast registration is more complex than normal IPv4 or
2578 * IPv6 multicast registration. Each Host Channel Adapter must register
2579 * with the Subnet Manager when it wishes to join a multicast group. It
2580 * should do so only once regardless of how many queue pairs it subscribes
2581 * to this group. And it should leave the group only after all queue pairs
2582 * attached to the group have been detached.
2584 * Return: true if the port must undertake the additional adminstrative
2585 * overhead of registering/unregistering with the SM and tracking of the
2586 * total number of queue pairs attached to the multicast group.
2588 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2590 return rdma_cap_ib_sa(device, port_num);
2594 * rdma_cap_af_ib - Check if the port of device has the capability
2595 * Native Infiniband Address.
2596 * @device: Device to check
2597 * @port_num: Port number to check
2599 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2600 * GID. RoCE uses a different mechanism, but still generates a GID via
2601 * a prescribed mechanism and port specific data.
2603 * Return: true if the port uses a GID address to identify devices on the
2606 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2608 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2612 * rdma_cap_eth_ah - Check if the port of device has the capability
2613 * Ethernet Address Handle.
2614 * @device: Device to check
2615 * @port_num: Port number to check
2617 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2618 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2619 * port. Normally, packet headers are generated by the sending host
2620 * adapter, but when sending connectionless datagrams, we must manually
2621 * inject the proper headers for the fabric we are communicating over.
2623 * Return: true if we are running as a RoCE port and must force the
2624 * addition of a Global Route Header built from our Ethernet Address
2625 * Handle into our header list for connectionless packets.
2627 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2629 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2633 * rdma_cap_opa_ah - Check if the port of device supports
2634 * OPA Address handles
2635 * @device: Device to check
2636 * @port_num: Port number to check
2638 * Return: true if we are running on an OPA device which supports
2639 * the extended OPA addressing.
2641 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2643 return (device->port_immutable[port_num].core_cap_flags &
2644 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2648 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2651 * @port_num: Port number
2653 * This MAD size includes the MAD headers and MAD payload. No other headers
2656 * Return the max MAD size required by the Port. Will return 0 if the port
2657 * does not support MADs
2659 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2661 return device->port_immutable[port_num].max_mad_size;
2665 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2666 * @device: Device to check
2667 * @port_num: Port number to check
2669 * RoCE GID table mechanism manages the various GIDs for a device.
2671 * NOTE: if allocating the port's GID table has failed, this call will still
2672 * return true, but any RoCE GID table API will fail.
2674 * Return: true if the port uses RoCE GID table mechanism in order to manage
2677 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2680 return rdma_protocol_roce(device, port_num) &&
2681 device->add_gid && device->del_gid;
2685 * Check if the device supports READ W/ INVALIDATE.
2687 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2690 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2691 * has support for it yet.
2693 return rdma_protocol_iwarp(dev, port_num);
2696 int ib_query_gid(struct ib_device *device,
2697 u8 port_num, int index, union ib_gid *gid,
2698 struct ib_gid_attr *attr);
2700 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2702 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2703 struct ifla_vf_info *info);
2704 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2705 struct ifla_vf_stats *stats);
2706 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2709 int ib_query_pkey(struct ib_device *device,
2710 u8 port_num, u16 index, u16 *pkey);
2712 int ib_modify_device(struct ib_device *device,
2713 int device_modify_mask,
2714 struct ib_device_modify *device_modify);
2716 int ib_modify_port(struct ib_device *device,
2717 u8 port_num, int port_modify_mask,
2718 struct ib_port_modify *port_modify);
2720 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2721 enum ib_gid_type gid_type, struct net_device *ndev,
2722 u8 *port_num, u16 *index);
2724 int ib_find_pkey(struct ib_device *device,
2725 u8 port_num, u16 pkey, u16 *index);
2729 * Create a memory registration for all memory in the system and place
2730 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2731 * ULPs to avoid the overhead of dynamic MRs.
2733 * This flag is generally considered unsafe and must only be used in
2734 * extremly trusted environments. Every use of it will log a warning
2735 * in the kernel log.
2737 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2740 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2741 const char *caller);
2742 #define ib_alloc_pd(device, flags) \
2743 __ib_alloc_pd((device), (flags), __func__)
2744 void ib_dealloc_pd(struct ib_pd *pd);
2747 * rdma_create_ah - Creates an address handle for the given address vector.
2748 * @pd: The protection domain associated with the address handle.
2749 * @ah_attr: The attributes of the address vector.
2751 * The address handle is used to reference a local or global destination
2752 * in all UD QP post sends.
2754 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2757 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2759 * @hdr: the L3 header to parse
2760 * @net_type: type of header to parse
2761 * @sgid: place to store source gid
2762 * @dgid: place to store destination gid
2764 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2765 enum rdma_network_type net_type,
2766 union ib_gid *sgid, union ib_gid *dgid);
2769 * ib_get_rdma_header_version - Get the header version
2770 * @hdr: the L3 header to parse
2772 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2775 * ib_init_ah_from_wc - Initializes address handle attributes from a
2777 * @device: Device on which the received message arrived.
2778 * @port_num: Port on which the received message arrived.
2779 * @wc: Work completion associated with the received message.
2780 * @grh: References the received global route header. This parameter is
2781 * ignored unless the work completion indicates that the GRH is valid.
2782 * @ah_attr: Returned attributes that can be used when creating an address
2783 * handle for replying to the message.
2785 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2786 const struct ib_wc *wc, const struct ib_grh *grh,
2787 struct rdma_ah_attr *ah_attr);
2790 * ib_create_ah_from_wc - Creates an address handle associated with the
2791 * sender of the specified work completion.
2792 * @pd: The protection domain associated with the address handle.
2793 * @wc: Work completion information associated with a received message.
2794 * @grh: References the received global route header. This parameter is
2795 * ignored unless the work completion indicates that the GRH is valid.
2796 * @port_num: The outbound port number to associate with the address.
2798 * The address handle is used to reference a local or global destination
2799 * in all UD QP post sends.
2801 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2802 const struct ib_grh *grh, u8 port_num);
2805 * rdma_modify_ah - Modifies the address vector associated with an address
2807 * @ah: The address handle to modify.
2808 * @ah_attr: The new address vector attributes to associate with the
2811 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2814 * rdma_query_ah - Queries the address vector associated with an address
2816 * @ah: The address handle to query.
2817 * @ah_attr: The address vector attributes associated with the address
2820 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2823 * rdma_destroy_ah - Destroys an address handle.
2824 * @ah: The address handle to destroy.
2826 int rdma_destroy_ah(struct ib_ah *ah);
2829 * ib_create_srq - Creates a SRQ associated with the specified protection
2831 * @pd: The protection domain associated with the SRQ.
2832 * @srq_init_attr: A list of initial attributes required to create the
2833 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2834 * the actual capabilities of the created SRQ.
2836 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2837 * requested size of the SRQ, and set to the actual values allocated
2838 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2839 * will always be at least as large as the requested values.
2841 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2842 struct ib_srq_init_attr *srq_init_attr);
2845 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2846 * @srq: The SRQ to modify.
2847 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2848 * the current values of selected SRQ attributes are returned.
2849 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2850 * are being modified.
2852 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2853 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2854 * the number of receives queued drops below the limit.
2856 int ib_modify_srq(struct ib_srq *srq,
2857 struct ib_srq_attr *srq_attr,
2858 enum ib_srq_attr_mask srq_attr_mask);
2861 * ib_query_srq - Returns the attribute list and current values for the
2863 * @srq: The SRQ to query.
2864 * @srq_attr: The attributes of the specified SRQ.
2866 int ib_query_srq(struct ib_srq *srq,
2867 struct ib_srq_attr *srq_attr);
2870 * ib_destroy_srq - Destroys the specified SRQ.
2871 * @srq: The SRQ to destroy.
2873 int ib_destroy_srq(struct ib_srq *srq);
2876 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2877 * @srq: The SRQ to post the work request on.
2878 * @recv_wr: A list of work requests to post on the receive queue.
2879 * @bad_recv_wr: On an immediate failure, this parameter will reference
2880 * the work request that failed to be posted on the QP.
2882 static inline int ib_post_srq_recv(struct ib_srq *srq,
2883 struct ib_recv_wr *recv_wr,
2884 struct ib_recv_wr **bad_recv_wr)
2886 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2890 * ib_create_qp - Creates a QP associated with the specified protection
2892 * @pd: The protection domain associated with the QP.
2893 * @qp_init_attr: A list of initial attributes required to create the
2894 * QP. If QP creation succeeds, then the attributes are updated to
2895 * the actual capabilities of the created QP.
2897 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2898 struct ib_qp_init_attr *qp_init_attr);
2901 * ib_modify_qp - Modifies the attributes for the specified QP and then
2902 * transitions the QP to the given state.
2903 * @qp: The QP to modify.
2904 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2905 * the current values of selected QP attributes are returned.
2906 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2907 * are being modified.
2909 int ib_modify_qp(struct ib_qp *qp,
2910 struct ib_qp_attr *qp_attr,
2914 * ib_query_qp - Returns the attribute list and current values for the
2916 * @qp: The QP to query.
2917 * @qp_attr: The attributes of the specified QP.
2918 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2919 * @qp_init_attr: Additional attributes of the selected QP.
2921 * The qp_attr_mask may be used to limit the query to gathering only the
2922 * selected attributes.
2924 int ib_query_qp(struct ib_qp *qp,
2925 struct ib_qp_attr *qp_attr,
2927 struct ib_qp_init_attr *qp_init_attr);
2930 * ib_destroy_qp - Destroys the specified QP.
2931 * @qp: The QP to destroy.
2933 int ib_destroy_qp(struct ib_qp *qp);
2936 * ib_open_qp - Obtain a reference to an existing sharable QP.
2937 * @xrcd - XRC domain
2938 * @qp_open_attr: Attributes identifying the QP to open.
2940 * Returns a reference to a sharable QP.
2942 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2943 struct ib_qp_open_attr *qp_open_attr);
2946 * ib_close_qp - Release an external reference to a QP.
2947 * @qp: The QP handle to release
2949 * The opened QP handle is released by the caller. The underlying
2950 * shared QP is not destroyed until all internal references are released.
2952 int ib_close_qp(struct ib_qp *qp);
2955 * ib_post_send - Posts a list of work requests to the send queue of
2957 * @qp: The QP to post the work request on.
2958 * @send_wr: A list of work requests to post on the send queue.
2959 * @bad_send_wr: On an immediate failure, this parameter will reference
2960 * the work request that failed to be posted on the QP.
2962 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2963 * error is returned, the QP state shall not be affected,
2964 * ib_post_send() will return an immediate error after queueing any
2965 * earlier work requests in the list.
2967 static inline int ib_post_send(struct ib_qp *qp,
2968 struct ib_send_wr *send_wr,
2969 struct ib_send_wr **bad_send_wr)
2971 return qp->device->post_send(qp, send_wr, bad_send_wr);
2975 * ib_post_recv - Posts a list of work requests to the receive queue of
2977 * @qp: The QP to post the work request on.
2978 * @recv_wr: A list of work requests to post on the receive queue.
2979 * @bad_recv_wr: On an immediate failure, this parameter will reference
2980 * the work request that failed to be posted on the QP.
2982 static inline int ib_post_recv(struct ib_qp *qp,
2983 struct ib_recv_wr *recv_wr,
2984 struct ib_recv_wr **bad_recv_wr)
2986 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2989 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2990 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2991 void ib_free_cq(struct ib_cq *cq);
2992 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2995 * ib_create_cq - Creates a CQ on the specified device.
2996 * @device: The device on which to create the CQ.
2997 * @comp_handler: A user-specified callback that is invoked when a
2998 * completion event occurs on the CQ.
2999 * @event_handler: A user-specified callback that is invoked when an
3000 * asynchronous event not associated with a completion occurs on the CQ.
3001 * @cq_context: Context associated with the CQ returned to the user via
3002 * the associated completion and event handlers.
3003 * @cq_attr: The attributes the CQ should be created upon.
3005 * Users can examine the cq structure to determine the actual CQ size.
3007 struct ib_cq *ib_create_cq(struct ib_device *device,
3008 ib_comp_handler comp_handler,
3009 void (*event_handler)(struct ib_event *, void *),
3011 const struct ib_cq_init_attr *cq_attr);
3014 * ib_resize_cq - Modifies the capacity of the CQ.
3015 * @cq: The CQ to resize.
3016 * @cqe: The minimum size of the CQ.
3018 * Users can examine the cq structure to determine the actual CQ size.
3020 int ib_resize_cq(struct ib_cq *cq, int cqe);
3023 * ib_modify_cq - Modifies moderation params of the CQ
3024 * @cq: The CQ to modify.
3025 * @cq_count: number of CQEs that will trigger an event
3026 * @cq_period: max period of time in usec before triggering an event
3029 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3032 * ib_destroy_cq - Destroys the specified CQ.
3033 * @cq: The CQ to destroy.
3035 int ib_destroy_cq(struct ib_cq *cq);
3038 * ib_poll_cq - poll a CQ for completion(s)
3039 * @cq:the CQ being polled
3040 * @num_entries:maximum number of completions to return
3041 * @wc:array of at least @num_entries &struct ib_wc where completions
3044 * Poll a CQ for (possibly multiple) completions. If the return value
3045 * is < 0, an error occurred. If the return value is >= 0, it is the
3046 * number of completions returned. If the return value is
3047 * non-negative and < num_entries, then the CQ was emptied.
3049 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3052 return cq->device->poll_cq(cq, num_entries, wc);
3056 * ib_peek_cq - Returns the number of unreaped completions currently
3057 * on the specified CQ.
3058 * @cq: The CQ to peek.
3059 * @wc_cnt: A minimum number of unreaped completions to check for.
3061 * If the number of unreaped completions is greater than or equal to wc_cnt,
3062 * this function returns wc_cnt, otherwise, it returns the actual number of
3063 * unreaped completions.
3065 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3068 * ib_req_notify_cq - Request completion notification on a CQ.
3069 * @cq: The CQ to generate an event for.
3071 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3072 * to request an event on the next solicited event or next work
3073 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3074 * may also be |ed in to request a hint about missed events, as
3078 * < 0 means an error occurred while requesting notification
3079 * == 0 means notification was requested successfully, and if
3080 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3081 * were missed and it is safe to wait for another event. In
3082 * this case is it guaranteed that any work completions added
3083 * to the CQ since the last CQ poll will trigger a completion
3084 * notification event.
3085 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3086 * in. It means that the consumer must poll the CQ again to
3087 * make sure it is empty to avoid missing an event because of a
3088 * race between requesting notification and an entry being
3089 * added to the CQ. This return value means it is possible
3090 * (but not guaranteed) that a work completion has been added
3091 * to the CQ since the last poll without triggering a
3092 * completion notification event.
3094 static inline int ib_req_notify_cq(struct ib_cq *cq,
3095 enum ib_cq_notify_flags flags)
3097 return cq->device->req_notify_cq(cq, flags);
3101 * ib_req_ncomp_notif - Request completion notification when there are
3102 * at least the specified number of unreaped completions on the CQ.
3103 * @cq: The CQ to generate an event for.
3104 * @wc_cnt: The number of unreaped completions that should be on the
3105 * CQ before an event is generated.
3107 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3109 return cq->device->req_ncomp_notif ?
3110 cq->device->req_ncomp_notif(cq, wc_cnt) :
3115 * ib_dma_mapping_error - check a DMA addr for error
3116 * @dev: The device for which the dma_addr was created
3117 * @dma_addr: The DMA address to check
3119 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3121 return dma_mapping_error(dev->dma_device, dma_addr);
3125 * ib_dma_map_single - Map a kernel virtual address to DMA address
3126 * @dev: The device for which the dma_addr is to be created
3127 * @cpu_addr: The kernel virtual address
3128 * @size: The size of the region in bytes
3129 * @direction: The direction of the DMA
3131 static inline u64 ib_dma_map_single(struct ib_device *dev,
3132 void *cpu_addr, size_t size,
3133 enum dma_data_direction direction)
3135 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3139 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3140 * @dev: The device for which the DMA address was created
3141 * @addr: The DMA address
3142 * @size: The size of the region in bytes
3143 * @direction: The direction of the DMA
3145 static inline void ib_dma_unmap_single(struct ib_device *dev,
3146 u64 addr, size_t size,
3147 enum dma_data_direction direction)
3149 dma_unmap_single(dev->dma_device, addr, size, direction);
3153 * ib_dma_map_page - Map a physical page to DMA address
3154 * @dev: The device for which the dma_addr is to be created
3155 * @page: The page to be mapped
3156 * @offset: The offset within the page
3157 * @size: The size of the region in bytes
3158 * @direction: The direction of the DMA
3160 static inline u64 ib_dma_map_page(struct ib_device *dev,
3162 unsigned long offset,
3164 enum dma_data_direction direction)
3166 return dma_map_page(dev->dma_device, page, offset, size, direction);
3170 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3171 * @dev: The device for which the DMA address was created
3172 * @addr: The DMA address
3173 * @size: The size of the region in bytes
3174 * @direction: The direction of the DMA
3176 static inline void ib_dma_unmap_page(struct ib_device *dev,
3177 u64 addr, size_t size,
3178 enum dma_data_direction direction)
3180 dma_unmap_page(dev->dma_device, addr, size, direction);
3184 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3185 * @dev: The device for which the DMA addresses are to be created
3186 * @sg: The array of scatter/gather entries
3187 * @nents: The number of scatter/gather entries
3188 * @direction: The direction of the DMA
3190 static inline int ib_dma_map_sg(struct ib_device *dev,
3191 struct scatterlist *sg, int nents,
3192 enum dma_data_direction direction)
3194 return dma_map_sg(dev->dma_device, sg, nents, direction);
3198 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3199 * @dev: The device for which the DMA addresses were created
3200 * @sg: The array of scatter/gather entries
3201 * @nents: The number of scatter/gather entries
3202 * @direction: The direction of the DMA
3204 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3205 struct scatterlist *sg, int nents,
3206 enum dma_data_direction direction)
3208 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3211 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3212 struct scatterlist *sg, int nents,
3213 enum dma_data_direction direction,
3214 unsigned long dma_attrs)
3216 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3220 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3221 struct scatterlist *sg, int nents,
3222 enum dma_data_direction direction,
3223 unsigned long dma_attrs)
3225 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3228 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3229 * @dev: The device for which the DMA addresses were created
3230 * @sg: The scatter/gather entry
3232 * Note: this function is obsolete. To do: change all occurrences of
3233 * ib_sg_dma_address() into sg_dma_address().
3235 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3236 struct scatterlist *sg)
3238 return sg_dma_address(sg);
3242 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3243 * @dev: The device for which the DMA addresses were created
3244 * @sg: The scatter/gather entry
3246 * Note: this function is obsolete. To do: change all occurrences of
3247 * ib_sg_dma_len() into sg_dma_len().
3249 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3250 struct scatterlist *sg)
3252 return sg_dma_len(sg);
3256 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3257 * @dev: The device for which the DMA address was created
3258 * @addr: The DMA address
3259 * @size: The size of the region in bytes
3260 * @dir: The direction of the DMA
3262 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3265 enum dma_data_direction dir)
3267 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3271 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3272 * @dev: The device for which the DMA address was created
3273 * @addr: The DMA address
3274 * @size: The size of the region in bytes
3275 * @dir: The direction of the DMA
3277 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3280 enum dma_data_direction dir)
3282 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3286 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3287 * @dev: The device for which the DMA address is requested
3288 * @size: The size of the region to allocate in bytes
3289 * @dma_handle: A pointer for returning the DMA address of the region
3290 * @flag: memory allocator flags
3292 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3294 dma_addr_t *dma_handle,
3297 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3301 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3302 * @dev: The device for which the DMA addresses were allocated
3303 * @size: The size of the region
3304 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3305 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3307 static inline void ib_dma_free_coherent(struct ib_device *dev,
3308 size_t size, void *cpu_addr,
3309 dma_addr_t dma_handle)
3311 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3315 * ib_dereg_mr - Deregisters a memory region and removes it from the
3316 * HCA translation table.
3317 * @mr: The memory region to deregister.
3319 * This function can fail, if the memory region has memory windows bound to it.
3321 int ib_dereg_mr(struct ib_mr *mr);
3323 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3324 enum ib_mr_type mr_type,
3328 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3330 * @mr - struct ib_mr pointer to be updated.
3331 * @newkey - new key to be used.
3333 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3335 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3336 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3340 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3341 * for calculating a new rkey for type 2 memory windows.
3342 * @rkey - the rkey to increment.
3344 static inline u32 ib_inc_rkey(u32 rkey)
3346 const u32 mask = 0x000000ff;
3347 return ((rkey + 1) & mask) | (rkey & ~mask);
3351 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3352 * @pd: The protection domain associated with the unmapped region.
3353 * @mr_access_flags: Specifies the memory access rights.
3354 * @fmr_attr: Attributes of the unmapped region.
3356 * A fast memory region must be mapped before it can be used as part of
3359 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3360 int mr_access_flags,
3361 struct ib_fmr_attr *fmr_attr);
3364 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3365 * @fmr: The fast memory region to associate with the pages.
3366 * @page_list: An array of physical pages to map to the fast memory region.
3367 * @list_len: The number of pages in page_list.
3368 * @iova: The I/O virtual address to use with the mapped region.
3370 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3371 u64 *page_list, int list_len,
3374 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3378 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3379 * @fmr_list: A linked list of fast memory regions to unmap.
3381 int ib_unmap_fmr(struct list_head *fmr_list);
3384 * ib_dealloc_fmr - Deallocates a fast memory region.
3385 * @fmr: The fast memory region to deallocate.
3387 int ib_dealloc_fmr(struct ib_fmr *fmr);
3390 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3391 * @qp: QP to attach to the multicast group. The QP must be type
3393 * @gid: Multicast group GID.
3394 * @lid: Multicast group LID in host byte order.
3396 * In order to send and receive multicast packets, subnet
3397 * administration must have created the multicast group and configured
3398 * the fabric appropriately. The port associated with the specified
3399 * QP must also be a member of the multicast group.
3401 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3404 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3405 * @qp: QP to detach from the multicast group.
3406 * @gid: Multicast group GID.
3407 * @lid: Multicast group LID in host byte order.
3409 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3412 * ib_alloc_xrcd - Allocates an XRC domain.
3413 * @device: The device on which to allocate the XRC domain.
3415 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3418 * ib_dealloc_xrcd - Deallocates an XRC domain.
3419 * @xrcd: The XRC domain to deallocate.
3421 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3423 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3424 struct ib_flow_attr *flow_attr, int domain);
3425 int ib_destroy_flow(struct ib_flow *flow_id);
3427 static inline int ib_check_mr_access(int flags)
3430 * Local write permission is required if remote write or
3431 * remote atomic permission is also requested.
3433 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3434 !(flags & IB_ACCESS_LOCAL_WRITE))
3441 * ib_check_mr_status: lightweight check of MR status.
3442 * This routine may provide status checks on a selected
3443 * ib_mr. first use is for signature status check.
3445 * @mr: A memory region.
3446 * @check_mask: Bitmask of which checks to perform from
3447 * ib_mr_status_check enumeration.
3448 * @mr_status: The container of relevant status checks.
3449 * failed checks will be indicated in the status bitmask
3450 * and the relevant info shall be in the error item.
3452 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3453 struct ib_mr_status *mr_status);
3455 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3456 u16 pkey, const union ib_gid *gid,
3457 const struct sockaddr *addr);
3458 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3459 struct ib_wq_init_attr *init_attr);
3460 int ib_destroy_wq(struct ib_wq *wq);
3461 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3463 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3464 struct ib_rwq_ind_table_init_attr*
3465 wq_ind_table_init_attr);
3466 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3468 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3469 unsigned int *sg_offset, unsigned int page_size);
3472 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3473 unsigned int *sg_offset, unsigned int page_size)
3477 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3483 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3484 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3486 void ib_drain_rq(struct ib_qp *qp);
3487 void ib_drain_sq(struct ib_qp *qp);
3488 void ib_drain_qp(struct ib_qp *qp);
3490 int ib_resolve_eth_dmac(struct ib_device *device,
3491 struct rdma_ah_attr *ah_attr);
3493 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3495 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3496 return attr->roce.dmac;
3500 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3502 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3503 attr->ib.dlid = (u16)dlid;
3504 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3505 attr->opa.dlid = dlid;
3508 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3510 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3511 return attr->ib.dlid;
3512 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3513 return attr->opa.dlid;
3517 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3522 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3527 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3530 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3531 attr->ib.src_path_bits = src_path_bits;
3532 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3533 attr->opa.src_path_bits = src_path_bits;
3536 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3538 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3539 return attr->ib.src_path_bits;
3540 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3541 return attr->opa.src_path_bits;
3545 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3547 attr->port_num = port_num;
3550 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3552 return attr->port_num;
3555 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3558 attr->static_rate = static_rate;
3561 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3563 return attr->static_rate;
3566 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3567 enum ib_ah_flags flag)
3569 attr->ah_flags = flag;
3572 static inline enum ib_ah_flags
3573 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3575 return attr->ah_flags;
3578 static inline const struct ib_global_route
3579 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3584 /*To retrieve and modify the grh */
3585 static inline struct ib_global_route
3586 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3591 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3593 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3595 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3598 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3601 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3603 grh->dgid.global.subnet_prefix = prefix;
3606 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3609 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3611 grh->dgid.global.interface_id = if_id;
3614 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3615 union ib_gid *dgid, u32 flow_label,
3616 u8 sgid_index, u8 hop_limit,
3619 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3621 attr->ah_flags = IB_AH_GRH;
3624 grh->flow_label = flow_label;
3625 grh->sgid_index = sgid_index;
3626 grh->hop_limit = hop_limit;
3627 grh->traffic_class = traffic_class;
3631 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3634 if ((rdma_protocol_roce(dev, port_num)) ||
3635 (rdma_protocol_iwarp(dev, port_num)))
3636 return RDMA_AH_ATTR_TYPE_ROCE;
3637 else if ((rdma_protocol_ib(dev, port_num)) &&
3638 (rdma_cap_opa_ah(dev, port_num)))
3639 return RDMA_AH_ATTR_TYPE_OPA;
3641 return RDMA_AH_ATTR_TYPE_IB;
3643 #endif /* IB_VERBS_H */