2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
67 #include <uapi/rdma/rdma_user_ioctl.h>
68 #include <uapi/rdma/ib_user_ioctl_verbs.h>
70 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
72 extern struct workqueue_struct *ib_wq;
73 extern struct workqueue_struct *ib_comp_wq;
74 extern struct workqueue_struct *ib_comp_unbound_wq;
84 extern union ib_gid zgid;
87 /* If link layer is Ethernet, this is RoCE V1 */
90 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
94 #define ROCE_V2_UDP_DPORT 4791
96 struct net_device *ndev;
97 struct ib_device *device;
99 enum ib_gid_type gid_type;
104 enum rdma_node_type {
105 /* IB values map to NodeInfo:NodeType. */
115 /* set the local administered indication */
116 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
119 enum rdma_transport_type {
121 RDMA_TRANSPORT_IWARP,
122 RDMA_TRANSPORT_USNIC,
123 RDMA_TRANSPORT_USNIC_UDP
126 enum rdma_protocol_type {
130 RDMA_PROTOCOL_USNIC_UDP
133 __attribute_const__ enum rdma_transport_type
134 rdma_node_get_transport(enum rdma_node_type node_type);
136 enum rdma_network_type {
138 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
143 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
145 if (network_type == RDMA_NETWORK_IPV4 ||
146 network_type == RDMA_NETWORK_IPV6)
147 return IB_GID_TYPE_ROCE_UDP_ENCAP;
149 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
150 return IB_GID_TYPE_IB;
153 static inline enum rdma_network_type
154 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
156 if (attr->gid_type == IB_GID_TYPE_IB)
157 return RDMA_NETWORK_IB;
159 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
160 return RDMA_NETWORK_IPV4;
162 return RDMA_NETWORK_IPV6;
165 enum rdma_link_layer {
166 IB_LINK_LAYER_UNSPECIFIED,
167 IB_LINK_LAYER_INFINIBAND,
168 IB_LINK_LAYER_ETHERNET,
171 enum ib_device_cap_flags {
172 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
173 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
174 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
175 IB_DEVICE_RAW_MULTI = (1 << 3),
176 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
177 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
178 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
179 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
180 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
181 /* Not in use, former INIT_TYPE = (1 << 9),*/
182 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
183 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
184 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
185 IB_DEVICE_SRQ_RESIZE = (1 << 13),
186 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
189 * This device supports a per-device lkey or stag that can be
190 * used without performing a memory registration for the local
191 * memory. Note that ULPs should never check this flag, but
192 * instead of use the local_dma_lkey flag in the ib_pd structure,
193 * which will always contain a usable lkey.
195 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
196 /* Reserved, old SEND_W_INV = (1 << 16),*/
197 IB_DEVICE_MEM_WINDOW = (1 << 17),
199 * Devices should set IB_DEVICE_UD_IP_SUM if they support
200 * insertion of UDP and TCP checksum on outgoing UD IPoIB
201 * messages and can verify the validity of checksum for
202 * incoming messages. Setting this flag implies that the
203 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
205 IB_DEVICE_UD_IP_CSUM = (1 << 18),
206 IB_DEVICE_UD_TSO = (1 << 19),
207 IB_DEVICE_XRC = (1 << 20),
210 * This device supports the IB "base memory management extension",
211 * which includes support for fast registrations (IB_WR_REG_MR,
212 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
213 * also be set by any iWarp device which must support FRs to comply
214 * to the iWarp verbs spec. iWarp devices also support the
215 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
218 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
219 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
220 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
221 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
222 IB_DEVICE_RC_IP_CSUM = (1 << 25),
223 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
224 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
226 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
227 * support execution of WQEs that involve synchronization
228 * of I/O operations with single completion queue managed
231 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
232 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
233 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
234 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
235 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
236 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
237 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
238 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
239 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
240 /* The device supports padding incoming writes to cacheline. */
241 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
244 enum ib_signature_prot_cap {
245 IB_PROT_T10DIF_TYPE_1 = 1,
246 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
247 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
250 enum ib_signature_guard_cap {
251 IB_GUARD_T10DIF_CRC = 1,
252 IB_GUARD_T10DIF_CSUM = 1 << 1,
261 enum ib_odp_general_cap_bits {
262 IB_ODP_SUPPORT = 1 << 0,
263 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
266 enum ib_odp_transport_cap_bits {
267 IB_ODP_SUPPORT_SEND = 1 << 0,
268 IB_ODP_SUPPORT_RECV = 1 << 1,
269 IB_ODP_SUPPORT_WRITE = 1 << 2,
270 IB_ODP_SUPPORT_READ = 1 << 3,
271 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
275 uint64_t general_caps;
277 uint32_t rc_odp_caps;
278 uint32_t uc_odp_caps;
279 uint32_t ud_odp_caps;
280 } per_transport_caps;
284 /* Corresponding bit will be set if qp type from
285 * 'enum ib_qp_type' is supported, e.g.
286 * supported_qpts |= 1 << IB_QPT_UD
289 u32 max_rwq_indirection_tables;
290 u32 max_rwq_indirection_table_size;
293 enum ib_tm_cap_flags {
294 /* Support tag matching with rendezvous offload for RC transport */
295 IB_TM_CAP_RNDV_RC = 1 << 0,
299 /* Max size of RNDV header */
300 u32 max_rndv_hdr_size;
301 /* Max number of entries in tag matching list */
303 /* From enum ib_tm_cap_flags */
305 /* Max number of outstanding list operations */
307 /* Max number of SGE in tag matching entry */
311 struct ib_cq_init_attr {
317 enum ib_cq_attr_mask {
318 IB_CQ_MODERATE = 1 << 0,
322 u16 max_cq_moderation_count;
323 u16 max_cq_moderation_period;
326 struct ib_dm_mr_attr {
332 struct ib_dm_alloc_attr {
338 struct ib_device_attr {
340 __be64 sys_image_guid;
348 u64 device_cap_flags;
359 int max_qp_init_rd_atom;
360 int max_ee_init_rd_atom;
361 enum ib_atomic_cap atomic_cap;
362 enum ib_atomic_cap masked_atomic_cap;
369 int max_mcast_qp_attach;
370 int max_total_mcast_qp_attach;
377 unsigned int max_fast_reg_page_list_len;
379 u8 local_ca_ack_delay;
382 struct ib_odp_caps odp_caps;
383 uint64_t timestamp_mask;
384 uint64_t hca_core_clock; /* in KHZ */
385 struct ib_rss_caps rss_caps;
387 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
388 struct ib_tm_caps tm_caps;
389 struct ib_cq_caps cq_caps;
401 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
404 case IB_MTU_256: return 256;
405 case IB_MTU_512: return 512;
406 case IB_MTU_1024: return 1024;
407 case IB_MTU_2048: return 2048;
408 case IB_MTU_4096: return 4096;
413 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
417 else if (mtu >= 2048)
419 else if (mtu >= 1024)
433 IB_PORT_ACTIVE_DEFER = 5
443 static inline int ib_width_enum_to_int(enum ib_port_width width)
446 case IB_WIDTH_1X: return 1;
447 case IB_WIDTH_4X: return 4;
448 case IB_WIDTH_8X: return 8;
449 case IB_WIDTH_12X: return 12;
465 * struct rdma_hw_stats
466 * @lock - Mutex to protect parallel write access to lifespan and values
467 * of counters, which are 64bits and not guaranteeed to be written
468 * atomicaly on 32bits systems.
469 * @timestamp - Used by the core code to track when the last update was
470 * @lifespan - Used by the core code to determine how old the counters
471 * should be before being updated again. Stored in jiffies, defaults
472 * to 10 milliseconds, drivers can override the default be specifying
473 * their own value during their allocation routine.
474 * @name - Array of pointers to static names used for the counters in
476 * @num_counters - How many hardware counters there are. If name is
477 * shorter than this number, a kernel oops will result. Driver authors
478 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
479 * in their code to prevent this.
480 * @value - Array of u64 counters that are accessed by the sysfs code and
481 * filled in by the drivers get_stats routine
483 struct rdma_hw_stats {
484 struct mutex lock; /* Protect lifespan and values[] */
485 unsigned long timestamp;
486 unsigned long lifespan;
487 const char * const *names;
492 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
494 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
496 * @names - Array of static const char *
497 * @num_counters - How many elements in array
498 * @lifespan - How many milliseconds between updates
500 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
501 const char * const *names, int num_counters,
502 unsigned long lifespan)
504 struct rdma_hw_stats *stats;
506 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
510 stats->names = names;
511 stats->num_counters = num_counters;
512 stats->lifespan = msecs_to_jiffies(lifespan);
518 /* Define bits for the various functionality this port needs to be supported by
521 /* Management 0x00000FFF */
522 #define RDMA_CORE_CAP_IB_MAD 0x00000001
523 #define RDMA_CORE_CAP_IB_SMI 0x00000002
524 #define RDMA_CORE_CAP_IB_CM 0x00000004
525 #define RDMA_CORE_CAP_IW_CM 0x00000008
526 #define RDMA_CORE_CAP_IB_SA 0x00000010
527 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
529 /* Address format 0x000FF000 */
530 #define RDMA_CORE_CAP_AF_IB 0x00001000
531 #define RDMA_CORE_CAP_ETH_AH 0x00002000
532 #define RDMA_CORE_CAP_OPA_AH 0x00004000
533 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
535 /* Protocol 0xFFF00000 */
536 #define RDMA_CORE_CAP_PROT_IB 0x00100000
537 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
538 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
539 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
540 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
541 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
543 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
544 | RDMA_CORE_CAP_PROT_ROCE \
545 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
547 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
548 | RDMA_CORE_CAP_IB_MAD \
549 | RDMA_CORE_CAP_IB_SMI \
550 | RDMA_CORE_CAP_IB_CM \
551 | RDMA_CORE_CAP_IB_SA \
552 | RDMA_CORE_CAP_AF_IB)
553 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
554 | RDMA_CORE_CAP_IB_MAD \
555 | RDMA_CORE_CAP_IB_CM \
556 | RDMA_CORE_CAP_AF_IB \
557 | RDMA_CORE_CAP_ETH_AH)
558 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
559 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
560 | RDMA_CORE_CAP_IB_MAD \
561 | RDMA_CORE_CAP_IB_CM \
562 | RDMA_CORE_CAP_AF_IB \
563 | RDMA_CORE_CAP_ETH_AH)
564 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
565 | RDMA_CORE_CAP_IW_CM)
566 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
567 | RDMA_CORE_CAP_OPA_MAD)
569 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
571 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
573 struct ib_port_attr {
575 enum ib_port_state state;
577 enum ib_mtu active_mtu;
579 unsigned int ip_gids:1;
580 /* This is the value from PortInfo CapabilityMask, defined by IBA */
598 enum ib_device_modify_flags {
599 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
600 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
603 #define IB_DEVICE_NODE_DESC_MAX 64
605 struct ib_device_modify {
607 char node_desc[IB_DEVICE_NODE_DESC_MAX];
610 enum ib_port_modify_flags {
611 IB_PORT_SHUTDOWN = 1,
612 IB_PORT_INIT_TYPE = (1<<2),
613 IB_PORT_RESET_QKEY_CNTR = (1<<3),
614 IB_PORT_OPA_MASK_CHG = (1<<4)
617 struct ib_port_modify {
618 u32 set_port_cap_mask;
619 u32 clr_port_cap_mask;
627 IB_EVENT_QP_ACCESS_ERR,
631 IB_EVENT_PATH_MIG_ERR,
632 IB_EVENT_DEVICE_FATAL,
633 IB_EVENT_PORT_ACTIVE,
636 IB_EVENT_PKEY_CHANGE,
639 IB_EVENT_SRQ_LIMIT_REACHED,
640 IB_EVENT_QP_LAST_WQE_REACHED,
641 IB_EVENT_CLIENT_REREGISTER,
646 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
649 struct ib_device *device;
657 enum ib_event_type event;
660 struct ib_event_handler {
661 struct ib_device *device;
662 void (*handler)(struct ib_event_handler *, struct ib_event *);
663 struct list_head list;
666 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
668 (_ptr)->device = _device; \
669 (_ptr)->handler = _handler; \
670 INIT_LIST_HEAD(&(_ptr)->list); \
673 struct ib_global_route {
674 const struct ib_gid_attr *sgid_attr;
683 __be32 version_tclass_flow;
691 union rdma_network_hdr {
694 /* The IB spec states that if it's IPv4, the header
695 * is located in the last 20 bytes of the header.
698 struct iphdr roce4grh;
702 #define IB_QPN_MASK 0xFFFFFF
705 IB_MULTICAST_QPN = 0xffffff
708 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
709 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
716 IB_RATE_PORT_CURRENT = 0,
717 IB_RATE_2_5_GBPS = 2,
725 IB_RATE_120_GBPS = 10,
726 IB_RATE_14_GBPS = 11,
727 IB_RATE_56_GBPS = 12,
728 IB_RATE_112_GBPS = 13,
729 IB_RATE_168_GBPS = 14,
730 IB_RATE_25_GBPS = 15,
731 IB_RATE_100_GBPS = 16,
732 IB_RATE_200_GBPS = 17,
733 IB_RATE_300_GBPS = 18
737 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
738 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
739 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
740 * @rate: rate to convert.
742 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
745 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
746 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
747 * @rate: rate to convert.
749 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
753 * enum ib_mr_type - memory region type
754 * @IB_MR_TYPE_MEM_REG: memory region that is used for
755 * normal registration
756 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
757 * signature operations (data-integrity
759 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
760 * register any arbitrary sg lists (without
761 * the normal mr constraints - see
766 IB_MR_TYPE_SIGNATURE,
772 * IB_SIG_TYPE_NONE: Unprotected.
773 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
775 enum ib_signature_type {
781 * Signature T10-DIF block-guard types
782 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
783 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
785 enum ib_t10_dif_bg_type {
791 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
793 * @bg_type: T10-DIF block guard type (CRC|CSUM)
794 * @pi_interval: protection information interval.
795 * @bg: seed of guard computation.
796 * @app_tag: application tag of guard block
797 * @ref_tag: initial guard block reference tag.
798 * @ref_remap: Indicate wethear the reftag increments each block
799 * @app_escape: Indicate to skip block check if apptag=0xffff
800 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
801 * @apptag_check_mask: check bitmask of application tag.
803 struct ib_t10_dif_domain {
804 enum ib_t10_dif_bg_type bg_type;
812 u16 apptag_check_mask;
816 * struct ib_sig_domain - Parameters for signature domain
817 * @sig_type: specific signauture type
818 * @sig: union of all signature domain attributes that may
819 * be used to set domain layout.
821 struct ib_sig_domain {
822 enum ib_signature_type sig_type;
824 struct ib_t10_dif_domain dif;
829 * struct ib_sig_attrs - Parameters for signature handover operation
830 * @check_mask: bitmask for signature byte check (8 bytes)
831 * @mem: memory domain layout desciptor.
832 * @wire: wire domain layout desciptor.
834 struct ib_sig_attrs {
836 struct ib_sig_domain mem;
837 struct ib_sig_domain wire;
840 enum ib_sig_err_type {
847 * Signature check masks (8 bytes in total) according to the T10-PI standard:
848 * -------- -------- ------------
849 * | GUARD | APPTAG | REFTAG |
851 * -------- -------- ------------
854 IB_SIG_CHECK_GUARD = 0xc0,
855 IB_SIG_CHECK_APPTAG = 0x30,
856 IB_SIG_CHECK_REFTAG = 0x0f,
860 * struct ib_sig_err - signature error descriptor
863 enum ib_sig_err_type err_type;
870 enum ib_mr_status_check {
871 IB_MR_CHECK_SIG_STATUS = 1,
875 * struct ib_mr_status - Memory region status container
877 * @fail_status: Bitmask of MR checks status. For each
878 * failed check a corresponding status bit is set.
879 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
882 struct ib_mr_status {
884 struct ib_sig_err sig_err;
888 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
890 * @mult: multiple to convert.
892 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
894 enum rdma_ah_attr_type {
895 RDMA_AH_ATTR_TYPE_UNDEFINED,
896 RDMA_AH_ATTR_TYPE_IB,
897 RDMA_AH_ATTR_TYPE_ROCE,
898 RDMA_AH_ATTR_TYPE_OPA,
906 struct roce_ah_attr {
916 struct rdma_ah_attr {
917 struct ib_global_route grh;
922 enum rdma_ah_attr_type type;
924 struct ib_ah_attr ib;
925 struct roce_ah_attr roce;
926 struct opa_ah_attr opa;
934 IB_WC_LOC_EEC_OP_ERR,
939 IB_WC_LOC_ACCESS_ERR,
940 IB_WC_REM_INV_REQ_ERR,
941 IB_WC_REM_ACCESS_ERR,
944 IB_WC_RNR_RETRY_EXC_ERR,
945 IB_WC_LOC_RDD_VIOL_ERR,
946 IB_WC_REM_INV_RD_REQ_ERR,
949 IB_WC_INV_EEC_STATE_ERR,
951 IB_WC_RESP_TIMEOUT_ERR,
955 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
966 IB_WC_MASKED_COMP_SWAP,
967 IB_WC_MASKED_FETCH_ADD,
969 * Set value of IB_WC_RECV so consumers can test if a completion is a
970 * receive by testing (opcode & IB_WC_RECV).
973 IB_WC_RECV_RDMA_WITH_IMM
978 IB_WC_WITH_IMM = (1<<1),
979 IB_WC_WITH_INVALIDATE = (1<<2),
980 IB_WC_IP_CSUM_OK = (1<<3),
981 IB_WC_WITH_SMAC = (1<<4),
982 IB_WC_WITH_VLAN = (1<<5),
983 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
989 struct ib_cqe *wr_cqe;
991 enum ib_wc_status status;
992 enum ib_wc_opcode opcode;
1006 u8 port_num; /* valid only for DR SMPs on switches */
1009 u8 network_hdr_type;
1012 enum ib_cq_notify_flags {
1013 IB_CQ_SOLICITED = 1 << 0,
1014 IB_CQ_NEXT_COMP = 1 << 1,
1015 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1016 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1025 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1027 return srq_type == IB_SRQT_XRC ||
1028 srq_type == IB_SRQT_TM;
1031 enum ib_srq_attr_mask {
1032 IB_SRQ_MAX_WR = 1 << 0,
1033 IB_SRQ_LIMIT = 1 << 1,
1036 struct ib_srq_attr {
1042 struct ib_srq_init_attr {
1043 void (*event_handler)(struct ib_event *, void *);
1045 struct ib_srq_attr attr;
1046 enum ib_srq_type srq_type;
1052 struct ib_xrcd *xrcd;
1067 u32 max_inline_data;
1070 * Maximum number of rdma_rw_ctx structures in flight at a time.
1071 * ib_create_qp() will calculate the right amount of neededed WRs
1072 * and MRs based on this.
1084 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1085 * here (and in that order) since the MAD layer uses them as
1086 * indices into a 2-entry table.
1095 IB_QPT_RAW_ETHERTYPE,
1096 IB_QPT_RAW_PACKET = 8,
1100 IB_QPT_DRIVER = 0xFF,
1101 /* Reserve a range for qp types internal to the low level driver.
1102 * These qp types will not be visible at the IB core layer, so the
1103 * IB_QPT_MAX usages should not be affected in the core layer
1105 IB_QPT_RESERVED1 = 0x1000,
1117 enum ib_qp_create_flags {
1118 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1119 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1120 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1121 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1122 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1123 IB_QP_CREATE_NETIF_QP = 1 << 5,
1124 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1125 /* FREE = 1 << 7, */
1126 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1127 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1128 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1129 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1130 /* reserve bits 26-31 for low level drivers' internal use */
1131 IB_QP_CREATE_RESERVED_START = 1 << 26,
1132 IB_QP_CREATE_RESERVED_END = 1 << 31,
1136 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1137 * callback to destroy the passed in QP.
1140 struct ib_qp_init_attr {
1141 void (*event_handler)(struct ib_event *, void *);
1143 struct ib_cq *send_cq;
1144 struct ib_cq *recv_cq;
1146 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1147 struct ib_qp_cap cap;
1148 enum ib_sig_type sq_sig_type;
1149 enum ib_qp_type qp_type;
1153 * Only needed for special QP types, or when using the RW API.
1156 struct ib_rwq_ind_table *rwq_ind_tbl;
1160 struct ib_qp_open_attr {
1161 void (*event_handler)(struct ib_event *, void *);
1164 enum ib_qp_type qp_type;
1167 enum ib_rnr_timeout {
1168 IB_RNR_TIMER_655_36 = 0,
1169 IB_RNR_TIMER_000_01 = 1,
1170 IB_RNR_TIMER_000_02 = 2,
1171 IB_RNR_TIMER_000_03 = 3,
1172 IB_RNR_TIMER_000_04 = 4,
1173 IB_RNR_TIMER_000_06 = 5,
1174 IB_RNR_TIMER_000_08 = 6,
1175 IB_RNR_TIMER_000_12 = 7,
1176 IB_RNR_TIMER_000_16 = 8,
1177 IB_RNR_TIMER_000_24 = 9,
1178 IB_RNR_TIMER_000_32 = 10,
1179 IB_RNR_TIMER_000_48 = 11,
1180 IB_RNR_TIMER_000_64 = 12,
1181 IB_RNR_TIMER_000_96 = 13,
1182 IB_RNR_TIMER_001_28 = 14,
1183 IB_RNR_TIMER_001_92 = 15,
1184 IB_RNR_TIMER_002_56 = 16,
1185 IB_RNR_TIMER_003_84 = 17,
1186 IB_RNR_TIMER_005_12 = 18,
1187 IB_RNR_TIMER_007_68 = 19,
1188 IB_RNR_TIMER_010_24 = 20,
1189 IB_RNR_TIMER_015_36 = 21,
1190 IB_RNR_TIMER_020_48 = 22,
1191 IB_RNR_TIMER_030_72 = 23,
1192 IB_RNR_TIMER_040_96 = 24,
1193 IB_RNR_TIMER_061_44 = 25,
1194 IB_RNR_TIMER_081_92 = 26,
1195 IB_RNR_TIMER_122_88 = 27,
1196 IB_RNR_TIMER_163_84 = 28,
1197 IB_RNR_TIMER_245_76 = 29,
1198 IB_RNR_TIMER_327_68 = 30,
1199 IB_RNR_TIMER_491_52 = 31
1202 enum ib_qp_attr_mask {
1204 IB_QP_CUR_STATE = (1<<1),
1205 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1206 IB_QP_ACCESS_FLAGS = (1<<3),
1207 IB_QP_PKEY_INDEX = (1<<4),
1208 IB_QP_PORT = (1<<5),
1209 IB_QP_QKEY = (1<<6),
1211 IB_QP_PATH_MTU = (1<<8),
1212 IB_QP_TIMEOUT = (1<<9),
1213 IB_QP_RETRY_CNT = (1<<10),
1214 IB_QP_RNR_RETRY = (1<<11),
1215 IB_QP_RQ_PSN = (1<<12),
1216 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1217 IB_QP_ALT_PATH = (1<<14),
1218 IB_QP_MIN_RNR_TIMER = (1<<15),
1219 IB_QP_SQ_PSN = (1<<16),
1220 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1221 IB_QP_PATH_MIG_STATE = (1<<18),
1222 IB_QP_CAP = (1<<19),
1223 IB_QP_DEST_QPN = (1<<20),
1224 IB_QP_RESERVED1 = (1<<21),
1225 IB_QP_RESERVED2 = (1<<22),
1226 IB_QP_RESERVED3 = (1<<23),
1227 IB_QP_RESERVED4 = (1<<24),
1228 IB_QP_RATE_LIMIT = (1<<25),
1253 enum ib_qp_state qp_state;
1254 enum ib_qp_state cur_qp_state;
1255 enum ib_mtu path_mtu;
1256 enum ib_mig_state path_mig_state;
1261 int qp_access_flags;
1262 struct ib_qp_cap cap;
1263 struct rdma_ah_attr ah_attr;
1264 struct rdma_ah_attr alt_ah_attr;
1267 u8 en_sqd_async_notify;
1270 u8 max_dest_rd_atomic;
1282 /* These are shared with userspace */
1283 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1284 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1285 IB_WR_SEND = IB_UVERBS_WR_SEND,
1286 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1287 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1288 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1289 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1290 IB_WR_LSO = IB_UVERBS_WR_TSO,
1291 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1292 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1293 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1294 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1295 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1296 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1297 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1299 /* These are kernel only and can not be issued by userspace */
1300 IB_WR_REG_MR = 0x20,
1303 /* reserve values for low level drivers' internal use.
1304 * These values will not be used at all in the ib core layer.
1306 IB_WR_RESERVED1 = 0xf0,
1318 enum ib_send_flags {
1320 IB_SEND_SIGNALED = (1<<1),
1321 IB_SEND_SOLICITED = (1<<2),
1322 IB_SEND_INLINE = (1<<3),
1323 IB_SEND_IP_CSUM = (1<<4),
1325 /* reserve bits 26-31 for low level drivers' internal use */
1326 IB_SEND_RESERVED_START = (1 << 26),
1327 IB_SEND_RESERVED_END = (1 << 31),
1337 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1341 struct ib_send_wr *next;
1344 struct ib_cqe *wr_cqe;
1346 struct ib_sge *sg_list;
1348 enum ib_wr_opcode opcode;
1352 u32 invalidate_rkey;
1357 struct ib_send_wr wr;
1362 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1364 return container_of(wr, struct ib_rdma_wr, wr);
1367 struct ib_atomic_wr {
1368 struct ib_send_wr wr;
1372 u64 compare_add_mask;
1377 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1379 return container_of(wr, struct ib_atomic_wr, wr);
1383 struct ib_send_wr wr;
1390 u16 pkey_index; /* valid for GSI only */
1391 u8 port_num; /* valid for DR SMPs on switch only */
1394 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1396 return container_of(wr, struct ib_ud_wr, wr);
1400 struct ib_send_wr wr;
1406 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1408 return container_of(wr, struct ib_reg_wr, wr);
1411 struct ib_sig_handover_wr {
1412 struct ib_send_wr wr;
1413 struct ib_sig_attrs *sig_attrs;
1414 struct ib_mr *sig_mr;
1416 struct ib_sge *prot;
1419 static inline const struct ib_sig_handover_wr *
1420 sig_handover_wr(const struct ib_send_wr *wr)
1422 return container_of(wr, struct ib_sig_handover_wr, wr);
1426 struct ib_recv_wr *next;
1429 struct ib_cqe *wr_cqe;
1431 struct ib_sge *sg_list;
1435 enum ib_access_flags {
1436 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1437 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1438 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1439 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1440 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1441 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1442 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1443 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1445 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1449 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1450 * are hidden here instead of a uapi header!
1452 enum ib_mr_rereg_flags {
1453 IB_MR_REREG_TRANS = 1,
1454 IB_MR_REREG_PD = (1<<1),
1455 IB_MR_REREG_ACCESS = (1<<2),
1456 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1459 struct ib_fmr_attr {
1467 enum rdma_remove_reason {
1469 * Userspace requested uobject deletion or initial try
1470 * to remove uobject via cleanup. Call could fail
1472 RDMA_REMOVE_DESTROY,
1473 /* Context deletion. This call should delete the actual object itself */
1475 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1476 RDMA_REMOVE_DRIVER_REMOVE,
1477 /* uobj is being cleaned-up before being committed */
1481 struct ib_rdmacg_object {
1482 #ifdef CONFIG_CGROUP_RDMA
1483 struct rdma_cgroup *cg; /* owner rdma cgroup */
1487 struct ib_ucontext {
1488 struct ib_device *device;
1489 struct ib_uverbs_file *ufile;
1491 * 'closing' can be read by the driver only during a destroy callback,
1492 * it is set when we are closing the file descriptor and indicates
1493 * that mm_sem may be locked.
1497 bool cleanup_retryable;
1500 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1501 struct rb_root_cached umem_tree;
1503 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1504 * mmu notifiers registration.
1506 struct rw_semaphore umem_rwsem;
1507 void (*invalidate_range)(struct ib_umem *umem,
1508 unsigned long start, unsigned long end);
1510 struct mmu_notifier mn;
1511 atomic_t notifier_count;
1512 /* A list of umems that don't have private mmu notifier counters yet. */
1513 struct list_head no_private_counters;
1517 struct ib_rdmacg_object cg_obj;
1521 u64 user_handle; /* handle given to us by userspace */
1522 /* ufile & ucontext owning this object */
1523 struct ib_uverbs_file *ufile;
1524 /* FIXME, save memory: ufile->context == context */
1525 struct ib_ucontext *context; /* associated user context */
1526 void *object; /* containing object */
1527 struct list_head list; /* link to context's list */
1528 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1529 int id; /* index into kernel idr */
1531 atomic_t usecnt; /* protects exclusive access */
1532 struct rcu_head rcu; /* kfree_rcu() overhead */
1534 const struct uverbs_api_object *uapi_object;
1538 const void __user *inbuf;
1539 void __user *outbuf;
1547 struct ib_device *device;
1548 struct ib_uobject *uobject;
1549 atomic_t usecnt; /* count all resources */
1551 u32 unsafe_global_rkey;
1554 * Implementation details of the RDMA core, don't use in drivers:
1556 struct ib_mr *__internal_mr;
1557 struct rdma_restrack_entry res;
1561 struct ib_device *device;
1562 atomic_t usecnt; /* count all exposed resources */
1563 struct inode *inode;
1565 struct mutex tgt_qp_mutex;
1566 struct list_head tgt_qp_list;
1570 struct ib_device *device;
1572 struct ib_uobject *uobject;
1573 const struct ib_gid_attr *sgid_attr;
1574 enum rdma_ah_attr_type type;
1577 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1579 enum ib_poll_context {
1580 IB_POLL_DIRECT, /* caller context, no hw completions */
1581 IB_POLL_SOFTIRQ, /* poll from softirq context */
1582 IB_POLL_WORKQUEUE, /* poll from workqueue */
1583 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1587 struct ib_device *device;
1588 struct ib_uobject *uobject;
1589 ib_comp_handler comp_handler;
1590 void (*event_handler)(struct ib_event *, void *);
1593 atomic_t usecnt; /* count number of work queues */
1594 enum ib_poll_context poll_ctx;
1597 struct irq_poll iop;
1598 struct work_struct work;
1600 struct workqueue_struct *comp_wq;
1602 * Implementation details of the RDMA core, don't use in drivers:
1604 struct rdma_restrack_entry res;
1608 struct ib_device *device;
1610 struct ib_uobject *uobject;
1611 void (*event_handler)(struct ib_event *, void *);
1613 enum ib_srq_type srq_type;
1620 struct ib_xrcd *xrcd;
1627 enum ib_raw_packet_caps {
1628 /* Strip cvlan from incoming packet and report it in the matching work
1629 * completion is supported.
1631 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1632 /* Scatter FCS field of an incoming packet to host memory is supported.
1634 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1635 /* Checksum offloads are supported (for both send and receive). */
1636 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1637 /* When a packet is received for an RQ with no receive WQEs, the
1638 * packet processing is delayed.
1640 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1654 struct ib_device *device;
1655 struct ib_uobject *uobject;
1657 void (*event_handler)(struct ib_event *, void *);
1661 enum ib_wq_state state;
1662 enum ib_wq_type wq_type;
1667 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1668 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1669 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1670 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1673 struct ib_wq_init_attr {
1675 enum ib_wq_type wq_type;
1679 void (*event_handler)(struct ib_event *, void *);
1680 u32 create_flags; /* Use enum ib_wq_flags */
1683 enum ib_wq_attr_mask {
1684 IB_WQ_STATE = 1 << 0,
1685 IB_WQ_CUR_STATE = 1 << 1,
1686 IB_WQ_FLAGS = 1 << 2,
1690 enum ib_wq_state wq_state;
1691 enum ib_wq_state curr_wq_state;
1692 u32 flags; /* Use enum ib_wq_flags */
1693 u32 flags_mask; /* Use enum ib_wq_flags */
1696 struct ib_rwq_ind_table {
1697 struct ib_device *device;
1698 struct ib_uobject *uobject;
1701 u32 log_ind_tbl_size;
1702 struct ib_wq **ind_tbl;
1705 struct ib_rwq_ind_table_init_attr {
1706 u32 log_ind_tbl_size;
1707 /* Each entry is a pointer to Receive Work Queue */
1708 struct ib_wq **ind_tbl;
1711 enum port_pkey_state {
1712 IB_PORT_PKEY_NOT_VALID = 0,
1713 IB_PORT_PKEY_VALID = 1,
1714 IB_PORT_PKEY_LISTED = 2,
1717 struct ib_qp_security;
1719 struct ib_port_pkey {
1720 enum port_pkey_state state;
1723 struct list_head qp_list;
1724 struct list_head to_error_list;
1725 struct ib_qp_security *sec;
1728 struct ib_ports_pkeys {
1729 struct ib_port_pkey main;
1730 struct ib_port_pkey alt;
1733 struct ib_qp_security {
1735 struct ib_device *dev;
1736 /* Hold this mutex when changing port and pkey settings. */
1738 struct ib_ports_pkeys *ports_pkeys;
1739 /* A list of all open shared QP handles. Required to enforce security
1740 * properly for all users of a shared QP.
1742 struct list_head shared_qp_list;
1745 atomic_t error_list_count;
1746 struct completion error_complete;
1747 int error_comps_pending;
1751 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1752 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1755 struct ib_device *device;
1757 struct ib_cq *send_cq;
1758 struct ib_cq *recv_cq;
1761 struct list_head rdma_mrs;
1762 struct list_head sig_mrs;
1764 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1765 struct list_head xrcd_list;
1767 /* count times opened, mcast attaches, flow attaches */
1769 struct list_head open_list;
1770 struct ib_qp *real_qp;
1771 struct ib_uobject *uobject;
1772 void (*event_handler)(struct ib_event *, void *);
1774 /* sgid_attrs associated with the AV's */
1775 const struct ib_gid_attr *av_sgid_attr;
1776 const struct ib_gid_attr *alt_path_sgid_attr;
1780 enum ib_qp_type qp_type;
1781 struct ib_rwq_ind_table *rwq_ind_tbl;
1782 struct ib_qp_security *qp_sec;
1786 * Implementation details of the RDMA core, don't use in drivers:
1788 struct rdma_restrack_entry res;
1792 struct ib_device *device;
1795 struct ib_uobject *uobject;
1800 struct ib_device *device;
1806 unsigned int page_size;
1809 struct ib_uobject *uobject; /* user */
1810 struct list_head qp_entry; /* FR */
1816 * Implementation details of the RDMA core, don't use in drivers:
1818 struct rdma_restrack_entry res;
1822 struct ib_device *device;
1824 struct ib_uobject *uobject;
1826 enum ib_mw_type type;
1830 struct ib_device *device;
1832 struct list_head list;
1837 /* Supported steering options */
1838 enum ib_flow_attr_type {
1839 /* steering according to rule specifications */
1840 IB_FLOW_ATTR_NORMAL = 0x0,
1841 /* default unicast and multicast rule -
1842 * receive all Eth traffic which isn't steered to any QP
1844 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1845 /* default multicast rule -
1846 * receive all Eth multicast traffic which isn't steered to any QP
1848 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1849 /* sniffer rule - receive all port traffic */
1850 IB_FLOW_ATTR_SNIFFER = 0x3
1853 /* Supported steering header types */
1854 enum ib_flow_spec_type {
1856 IB_FLOW_SPEC_ETH = 0x20,
1857 IB_FLOW_SPEC_IB = 0x22,
1859 IB_FLOW_SPEC_IPV4 = 0x30,
1860 IB_FLOW_SPEC_IPV6 = 0x31,
1861 IB_FLOW_SPEC_ESP = 0x34,
1863 IB_FLOW_SPEC_TCP = 0x40,
1864 IB_FLOW_SPEC_UDP = 0x41,
1865 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1866 IB_FLOW_SPEC_GRE = 0x51,
1867 IB_FLOW_SPEC_MPLS = 0x60,
1868 IB_FLOW_SPEC_INNER = 0x100,
1870 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1871 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1872 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1873 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1875 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1876 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1878 /* Flow steering rule priority is set according to it's domain.
1879 * Lower domain value means higher priority.
1881 enum ib_flow_domain {
1882 IB_FLOW_DOMAIN_USER,
1883 IB_FLOW_DOMAIN_ETHTOOL,
1886 IB_FLOW_DOMAIN_NUM /* Must be last */
1889 enum ib_flow_flags {
1890 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1891 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1892 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1895 struct ib_flow_eth_filter {
1904 struct ib_flow_spec_eth {
1907 struct ib_flow_eth_filter val;
1908 struct ib_flow_eth_filter mask;
1911 struct ib_flow_ib_filter {
1918 struct ib_flow_spec_ib {
1921 struct ib_flow_ib_filter val;
1922 struct ib_flow_ib_filter mask;
1925 /* IPv4 header flags */
1926 enum ib_ipv4_flags {
1927 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1928 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1929 last have this flag set */
1932 struct ib_flow_ipv4_filter {
1943 struct ib_flow_spec_ipv4 {
1946 struct ib_flow_ipv4_filter val;
1947 struct ib_flow_ipv4_filter mask;
1950 struct ib_flow_ipv6_filter {
1961 struct ib_flow_spec_ipv6 {
1964 struct ib_flow_ipv6_filter val;
1965 struct ib_flow_ipv6_filter mask;
1968 struct ib_flow_tcp_udp_filter {
1975 struct ib_flow_spec_tcp_udp {
1978 struct ib_flow_tcp_udp_filter val;
1979 struct ib_flow_tcp_udp_filter mask;
1982 struct ib_flow_tunnel_filter {
1987 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1988 * the tunnel_id from val has the vni value
1990 struct ib_flow_spec_tunnel {
1993 struct ib_flow_tunnel_filter val;
1994 struct ib_flow_tunnel_filter mask;
1997 struct ib_flow_esp_filter {
2004 struct ib_flow_spec_esp {
2007 struct ib_flow_esp_filter val;
2008 struct ib_flow_esp_filter mask;
2011 struct ib_flow_gre_filter {
2012 __be16 c_ks_res0_ver;
2019 struct ib_flow_spec_gre {
2022 struct ib_flow_gre_filter val;
2023 struct ib_flow_gre_filter mask;
2026 struct ib_flow_mpls_filter {
2032 struct ib_flow_spec_mpls {
2035 struct ib_flow_mpls_filter val;
2036 struct ib_flow_mpls_filter mask;
2039 struct ib_flow_spec_action_tag {
2040 enum ib_flow_spec_type type;
2045 struct ib_flow_spec_action_drop {
2046 enum ib_flow_spec_type type;
2050 struct ib_flow_spec_action_handle {
2051 enum ib_flow_spec_type type;
2053 struct ib_flow_action *act;
2056 enum ib_counters_description {
2061 struct ib_flow_spec_action_count {
2062 enum ib_flow_spec_type type;
2064 struct ib_counters *counters;
2067 union ib_flow_spec {
2072 struct ib_flow_spec_eth eth;
2073 struct ib_flow_spec_ib ib;
2074 struct ib_flow_spec_ipv4 ipv4;
2075 struct ib_flow_spec_tcp_udp tcp_udp;
2076 struct ib_flow_spec_ipv6 ipv6;
2077 struct ib_flow_spec_tunnel tunnel;
2078 struct ib_flow_spec_esp esp;
2079 struct ib_flow_spec_gre gre;
2080 struct ib_flow_spec_mpls mpls;
2081 struct ib_flow_spec_action_tag flow_tag;
2082 struct ib_flow_spec_action_drop drop;
2083 struct ib_flow_spec_action_handle action;
2084 struct ib_flow_spec_action_count flow_count;
2087 struct ib_flow_attr {
2088 enum ib_flow_attr_type type;
2094 union ib_flow_spec flows[];
2099 struct ib_device *device;
2100 struct ib_uobject *uobject;
2103 enum ib_flow_action_type {
2104 IB_FLOW_ACTION_UNSPECIFIED,
2105 IB_FLOW_ACTION_ESP = 1,
2108 struct ib_flow_action_attrs_esp_keymats {
2109 enum ib_uverbs_flow_action_esp_keymat protocol;
2111 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2115 struct ib_flow_action_attrs_esp_replays {
2116 enum ib_uverbs_flow_action_esp_replay protocol;
2118 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2122 enum ib_flow_action_attrs_esp_flags {
2123 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2124 * This is done in order to share the same flags between user-space and
2125 * kernel and spare an unnecessary translation.
2129 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2130 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2133 struct ib_flow_spec_list {
2134 struct ib_flow_spec_list *next;
2135 union ib_flow_spec spec;
2138 struct ib_flow_action_attrs_esp {
2139 struct ib_flow_action_attrs_esp_keymats *keymat;
2140 struct ib_flow_action_attrs_esp_replays *replay;
2141 struct ib_flow_spec_list *encap;
2142 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2143 * Value of 0 is a valid value.
2149 /* Use enum ib_flow_action_attrs_esp_flags */
2151 u64 hard_limit_pkts;
2154 struct ib_flow_action {
2155 struct ib_device *device;
2156 struct ib_uobject *uobject;
2157 enum ib_flow_action_type type;
2164 enum ib_process_mad_flags {
2165 IB_MAD_IGNORE_MKEY = 1,
2166 IB_MAD_IGNORE_BKEY = 2,
2167 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2170 enum ib_mad_result {
2171 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2172 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2173 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2174 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2177 struct ib_port_cache {
2179 struct ib_pkey_cache *pkey;
2180 struct ib_gid_table *gid;
2182 enum ib_port_state port_state;
2187 struct ib_event_handler event_handler;
2188 struct ib_port_cache *ports;
2193 struct ib_port_immutable {
2200 /* rdma netdev type - specifies protocol type */
2201 enum rdma_netdev_t {
2202 RDMA_NETDEV_OPA_VNIC,
2207 * struct rdma_netdev - rdma netdev
2208 * For cases where netstack interfacing is required.
2210 struct rdma_netdev {
2212 struct ib_device *hca;
2216 * cleanup function must be specified.
2217 * FIXME: This is only used for OPA_VNIC and that usage should be
2220 void (*free_rdma_netdev)(struct net_device *netdev);
2222 /* control functions */
2223 void (*set_id)(struct net_device *netdev, int id);
2225 int (*send)(struct net_device *dev, struct sk_buff *skb,
2226 struct ib_ah *address, u32 dqpn);
2228 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2229 union ib_gid *gid, u16 mlid,
2230 int set_qkey, u32 qkey);
2231 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2232 union ib_gid *gid, u16 mlid);
2235 struct ib_port_pkey_list {
2236 /* Lock to hold while modifying the list. */
2237 spinlock_t list_lock;
2238 struct list_head pkey_list;
2241 struct ib_counters {
2242 struct ib_device *device;
2243 struct ib_uobject *uobject;
2244 /* num of objects attached */
2248 struct ib_counters_read_attr {
2251 u32 flags; /* use enum ib_read_counters_flags */
2254 struct uverbs_attr_bundle;
2257 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2258 struct device *dma_device;
2260 char name[IB_DEVICE_NAME_MAX];
2262 struct list_head event_handler_list;
2263 spinlock_t event_handler_lock;
2265 spinlock_t client_data_lock;
2266 struct list_head core_list;
2267 /* Access to the client_data_list is protected by the client_data_lock
2268 * spinlock and the lists_rwsem read-write semaphore */
2269 struct list_head client_data_list;
2271 struct ib_cache cache;
2273 * port_immutable is indexed by port number
2275 struct ib_port_immutable *port_immutable;
2277 int num_comp_vectors;
2279 struct ib_port_pkey_list *port_pkey_list;
2281 struct iw_cm_verbs *iwcm;
2284 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2285 * driver initialized data. The struct is kfree()'ed by the sysfs
2286 * core when the device is removed. A lifespan of -1 in the return
2287 * struct tells the core to set a default lifespan.
2289 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2292 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2293 * @index - The index in the value array we wish to have updated, or
2294 * num_counters if we want all stats updated
2296 * < 0 - Error, no counters updated
2297 * index - Updated the single counter pointed to by index
2298 * num_counters - Updated all counters (will reset the timestamp
2299 * and prevent further calls for lifespan milliseconds)
2300 * Drivers are allowed to update all counters in leiu of just the
2301 * one given in index at their option
2303 int (*get_hw_stats)(struct ib_device *device,
2304 struct rdma_hw_stats *stats,
2305 u8 port, int index);
2306 int (*query_device)(struct ib_device *device,
2307 struct ib_device_attr *device_attr,
2308 struct ib_udata *udata);
2309 int (*query_port)(struct ib_device *device,
2311 struct ib_port_attr *port_attr);
2312 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2314 /* When calling get_netdev, the HW vendor's driver should return the
2315 * net device of device @device at port @port_num or NULL if such
2316 * a net device doesn't exist. The vendor driver should call dev_hold
2317 * on this net device. The HW vendor's device driver must guarantee
2318 * that this function returns NULL before the net device has finished
2319 * NETDEV_UNREGISTER state.
2321 struct net_device *(*get_netdev)(struct ib_device *device,
2323 /* query_gid should be return GID value for @device, when @port_num
2324 * link layer is either IB or iWarp. It is no-op if @port_num port
2325 * is RoCE link layer.
2327 int (*query_gid)(struct ib_device *device,
2328 u8 port_num, int index,
2330 /* When calling add_gid, the HW vendor's driver should add the gid
2331 * of device of port at gid index available at @attr. Meta-info of
2332 * that gid (for example, the network device related to this gid) is
2333 * available at @attr. @context allows the HW vendor driver to store
2334 * extra information together with a GID entry. The HW vendor driver may
2335 * allocate memory to contain this information and store it in @context
2336 * when a new GID entry is written to. Params are consistent until the
2337 * next call of add_gid or delete_gid. The function should return 0 on
2338 * success or error otherwise. The function could be called
2339 * concurrently for different ports. This function is only called when
2340 * roce_gid_table is used.
2342 int (*add_gid)(const struct ib_gid_attr *attr,
2344 /* When calling del_gid, the HW vendor's driver should delete the
2345 * gid of device @device at gid index gid_index of port port_num
2346 * available in @attr.
2347 * Upon the deletion of a GID entry, the HW vendor must free any
2348 * allocated memory. The caller will clear @context afterwards.
2349 * This function is only called when roce_gid_table is used.
2351 int (*del_gid)(const struct ib_gid_attr *attr,
2353 int (*query_pkey)(struct ib_device *device,
2354 u8 port_num, u16 index, u16 *pkey);
2355 int (*modify_device)(struct ib_device *device,
2356 int device_modify_mask,
2357 struct ib_device_modify *device_modify);
2358 int (*modify_port)(struct ib_device *device,
2359 u8 port_num, int port_modify_mask,
2360 struct ib_port_modify *port_modify);
2361 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2362 struct ib_udata *udata);
2363 int (*dealloc_ucontext)(struct ib_ucontext *context);
2364 int (*mmap)(struct ib_ucontext *context,
2365 struct vm_area_struct *vma);
2366 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2367 struct ib_ucontext *context,
2368 struct ib_udata *udata);
2369 int (*dealloc_pd)(struct ib_pd *pd);
2370 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2371 struct rdma_ah_attr *ah_attr,
2372 struct ib_udata *udata);
2373 int (*modify_ah)(struct ib_ah *ah,
2374 struct rdma_ah_attr *ah_attr);
2375 int (*query_ah)(struct ib_ah *ah,
2376 struct rdma_ah_attr *ah_attr);
2377 int (*destroy_ah)(struct ib_ah *ah);
2378 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2379 struct ib_srq_init_attr *srq_init_attr,
2380 struct ib_udata *udata);
2381 int (*modify_srq)(struct ib_srq *srq,
2382 struct ib_srq_attr *srq_attr,
2383 enum ib_srq_attr_mask srq_attr_mask,
2384 struct ib_udata *udata);
2385 int (*query_srq)(struct ib_srq *srq,
2386 struct ib_srq_attr *srq_attr);
2387 int (*destroy_srq)(struct ib_srq *srq);
2388 int (*post_srq_recv)(struct ib_srq *srq,
2389 const struct ib_recv_wr *recv_wr,
2390 const struct ib_recv_wr **bad_recv_wr);
2391 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2392 struct ib_qp_init_attr *qp_init_attr,
2393 struct ib_udata *udata);
2394 int (*modify_qp)(struct ib_qp *qp,
2395 struct ib_qp_attr *qp_attr,
2397 struct ib_udata *udata);
2398 int (*query_qp)(struct ib_qp *qp,
2399 struct ib_qp_attr *qp_attr,
2401 struct ib_qp_init_attr *qp_init_attr);
2402 int (*destroy_qp)(struct ib_qp *qp);
2403 int (*post_send)(struct ib_qp *qp,
2404 const struct ib_send_wr *send_wr,
2405 const struct ib_send_wr **bad_send_wr);
2406 int (*post_recv)(struct ib_qp *qp,
2407 const struct ib_recv_wr *recv_wr,
2408 const struct ib_recv_wr **bad_recv_wr);
2409 struct ib_cq * (*create_cq)(struct ib_device *device,
2410 const struct ib_cq_init_attr *attr,
2411 struct ib_ucontext *context,
2412 struct ib_udata *udata);
2413 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2415 int (*destroy_cq)(struct ib_cq *cq);
2416 int (*resize_cq)(struct ib_cq *cq, int cqe,
2417 struct ib_udata *udata);
2418 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2420 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2421 int (*req_notify_cq)(struct ib_cq *cq,
2422 enum ib_cq_notify_flags flags);
2423 int (*req_ncomp_notif)(struct ib_cq *cq,
2425 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2426 int mr_access_flags);
2427 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2428 u64 start, u64 length,
2430 int mr_access_flags,
2431 struct ib_udata *udata);
2432 int (*rereg_user_mr)(struct ib_mr *mr,
2434 u64 start, u64 length,
2436 int mr_access_flags,
2438 struct ib_udata *udata);
2439 int (*dereg_mr)(struct ib_mr *mr);
2440 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2441 enum ib_mr_type mr_type,
2443 int (*map_mr_sg)(struct ib_mr *mr,
2444 struct scatterlist *sg,
2446 unsigned int *sg_offset);
2447 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2448 enum ib_mw_type type,
2449 struct ib_udata *udata);
2450 int (*dealloc_mw)(struct ib_mw *mw);
2451 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2452 int mr_access_flags,
2453 struct ib_fmr_attr *fmr_attr);
2454 int (*map_phys_fmr)(struct ib_fmr *fmr,
2455 u64 *page_list, int list_len,
2457 int (*unmap_fmr)(struct list_head *fmr_list);
2458 int (*dealloc_fmr)(struct ib_fmr *fmr);
2459 int (*attach_mcast)(struct ib_qp *qp,
2462 int (*detach_mcast)(struct ib_qp *qp,
2465 int (*process_mad)(struct ib_device *device,
2466 int process_mad_flags,
2468 const struct ib_wc *in_wc,
2469 const struct ib_grh *in_grh,
2470 const struct ib_mad_hdr *in_mad,
2472 struct ib_mad_hdr *out_mad,
2473 size_t *out_mad_size,
2474 u16 *out_mad_pkey_index);
2475 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2476 struct ib_ucontext *ucontext,
2477 struct ib_udata *udata);
2478 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2479 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2483 struct ib_udata *udata);
2484 int (*destroy_flow)(struct ib_flow *flow_id);
2485 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2486 struct ib_mr_status *mr_status);
2487 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2488 void (*drain_rq)(struct ib_qp *qp);
2489 void (*drain_sq)(struct ib_qp *qp);
2490 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2492 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2493 struct ifla_vf_info *ivf);
2494 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2495 struct ifla_vf_stats *stats);
2496 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2498 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2499 struct ib_wq_init_attr *init_attr,
2500 struct ib_udata *udata);
2501 int (*destroy_wq)(struct ib_wq *wq);
2502 int (*modify_wq)(struct ib_wq *wq,
2503 struct ib_wq_attr *attr,
2505 struct ib_udata *udata);
2506 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2507 struct ib_rwq_ind_table_init_attr *init_attr,
2508 struct ib_udata *udata);
2509 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2510 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2511 const struct ib_flow_action_attrs_esp *attr,
2512 struct uverbs_attr_bundle *attrs);
2513 int (*destroy_flow_action)(struct ib_flow_action *action);
2514 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2515 const struct ib_flow_action_attrs_esp *attr,
2516 struct uverbs_attr_bundle *attrs);
2517 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2518 struct ib_ucontext *context,
2519 struct ib_dm_alloc_attr *attr,
2520 struct uverbs_attr_bundle *attrs);
2521 int (*dealloc_dm)(struct ib_dm *dm);
2522 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2523 struct ib_dm_mr_attr *attr,
2524 struct uverbs_attr_bundle *attrs);
2525 struct ib_counters * (*create_counters)(struct ib_device *device,
2526 struct uverbs_attr_bundle *attrs);
2527 int (*destroy_counters)(struct ib_counters *counters);
2528 int (*read_counters)(struct ib_counters *counters,
2529 struct ib_counters_read_attr *counters_read_attr,
2530 struct uverbs_attr_bundle *attrs);
2533 * rdma netdev operation
2535 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2536 * doesn't support the specified rdma netdev type.
2538 struct net_device *(*alloc_rdma_netdev)(
2539 struct ib_device *device,
2541 enum rdma_netdev_t type,
2543 unsigned char name_assign_type,
2544 void (*setup)(struct net_device *));
2546 struct module *owner;
2548 struct kobject *ports_parent;
2549 struct list_head port_list;
2552 IB_DEV_UNINITIALIZED,
2558 u64 uverbs_cmd_mask;
2559 u64 uverbs_ex_cmd_mask;
2561 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2567 struct ib_device_attr attrs;
2568 struct attribute_group *hw_stats_ag;
2569 struct rdma_hw_stats *hw_stats;
2571 #ifdef CONFIG_CGROUP_RDMA
2572 struct rdmacg_device cg_device;
2577 * Implementation details of the RDMA core, don't use in drivers
2579 struct rdma_restrack_root res;
2582 * The following mandatory functions are used only at device
2583 * registration. Keep functions such as these at the end of this
2584 * structure to avoid cache line misses when accessing struct ib_device
2587 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2588 void (*get_dev_fw_str)(struct ib_device *, char *str);
2589 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2592 const struct uverbs_object_tree_def *const *driver_specs;
2593 enum rdma_driver_id driver_id;
2598 void (*add) (struct ib_device *);
2599 void (*remove)(struct ib_device *, void *client_data);
2601 /* Returns the net_dev belonging to this ib_client and matching the
2603 * @dev: An RDMA device that the net_dev use for communication.
2604 * @port: A physical port number on the RDMA device.
2605 * @pkey: P_Key that the net_dev uses if applicable.
2606 * @gid: A GID that the net_dev uses to communicate.
2607 * @addr: An IP address the net_dev is configured with.
2608 * @client_data: The device's client data set by ib_set_client_data().
2610 * An ib_client that implements a net_dev on top of RDMA devices
2611 * (such as IP over IB) should implement this callback, allowing the
2612 * rdma_cm module to find the right net_dev for a given request.
2614 * The caller is responsible for calling dev_put on the returned
2616 struct net_device *(*get_net_dev_by_params)(
2617 struct ib_device *dev,
2620 const union ib_gid *gid,
2621 const struct sockaddr *addr,
2623 struct list_head list;
2626 struct ib_device *ib_alloc_device(size_t size);
2627 void ib_dealloc_device(struct ib_device *device);
2629 void ib_get_device_fw_str(struct ib_device *device, char *str);
2631 int ib_register_device(struct ib_device *device,
2632 int (*port_callback)(struct ib_device *,
2633 u8, struct kobject *));
2634 void ib_unregister_device(struct ib_device *device);
2636 int ib_register_client (struct ib_client *client);
2637 void ib_unregister_client(struct ib_client *client);
2639 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2640 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2643 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2645 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2648 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2650 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2653 static inline bool ib_is_buffer_cleared(const void __user *p,
2659 if (len > USHRT_MAX)
2662 buf = memdup_user(p, len);
2666 ret = !memchr_inv(buf, 0, len);
2671 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2675 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2679 * ib_is_destroy_retryable - Check whether the uobject destruction
2681 * @ret: The initial destruction return code
2682 * @why: remove reason
2683 * @uobj: The uobject that is destroyed
2685 * This function is a helper function that IB layer and low-level drivers
2686 * can use to consider whether the destruction of the given uobject is
2688 * It checks the original return code, if it wasn't success the destruction
2689 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2690 * the remove reason. (i.e. why).
2691 * Must be called with the object locked for destroy.
2693 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2694 struct ib_uobject *uobj)
2696 return ret && (why == RDMA_REMOVE_DESTROY ||
2697 uobj->context->cleanup_retryable);
2701 * ib_destroy_usecnt - Called during destruction to check the usecnt
2702 * @usecnt: The usecnt atomic
2703 * @why: remove reason
2704 * @uobj: The uobject that is destroyed
2706 * Non-zero usecnts will block destruction unless destruction was triggered by
2707 * a ucontext cleanup.
2709 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2710 enum rdma_remove_reason why,
2711 struct ib_uobject *uobj)
2713 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2719 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2720 * contains all required attributes and no attributes not allowed for
2721 * the given QP state transition.
2722 * @cur_state: Current QP state
2723 * @next_state: Next QP state
2725 * @mask: Mask of supplied QP attributes
2726 * @ll : link layer of port
2728 * This function is a helper function that a low-level driver's
2729 * modify_qp method can use to validate the consumer's input. It
2730 * checks that cur_state and next_state are valid QP states, that a
2731 * transition from cur_state to next_state is allowed by the IB spec,
2732 * and that the attribute mask supplied is allowed for the transition.
2734 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2735 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2736 enum rdma_link_layer ll);
2738 void ib_register_event_handler(struct ib_event_handler *event_handler);
2739 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2740 void ib_dispatch_event(struct ib_event *event);
2742 int ib_query_port(struct ib_device *device,
2743 u8 port_num, struct ib_port_attr *port_attr);
2745 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2749 * rdma_cap_ib_switch - Check if the device is IB switch
2750 * @device: Device to check
2752 * Device driver is responsible for setting is_switch bit on
2753 * in ib_device structure at init time.
2755 * Return: true if the device is IB switch.
2757 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2759 return device->is_switch;
2763 * rdma_start_port - Return the first valid port number for the device
2766 * @device: Device to be checked
2768 * Return start port number
2770 static inline u8 rdma_start_port(const struct ib_device *device)
2772 return rdma_cap_ib_switch(device) ? 0 : 1;
2776 * rdma_end_port - Return the last valid port number for the device
2779 * @device: Device to be checked
2781 * Return last port number
2783 static inline u8 rdma_end_port(const struct ib_device *device)
2785 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2788 static inline int rdma_is_port_valid(const struct ib_device *device,
2791 return (port >= rdma_start_port(device) &&
2792 port <= rdma_end_port(device));
2795 static inline bool rdma_is_grh_required(const struct ib_device *device,
2798 return device->port_immutable[port_num].core_cap_flags &
2799 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2802 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2804 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2807 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2809 return device->port_immutable[port_num].core_cap_flags &
2810 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2813 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2815 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2818 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2823 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2825 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2828 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2830 return rdma_protocol_ib(device, port_num) ||
2831 rdma_protocol_roce(device, port_num);
2834 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2836 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2839 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2841 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2845 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2846 * Management Datagrams.
2847 * @device: Device to check
2848 * @port_num: Port number to check
2850 * Management Datagrams (MAD) are a required part of the InfiniBand
2851 * specification and are supported on all InfiniBand devices. A slightly
2852 * extended version are also supported on OPA interfaces.
2854 * Return: true if the port supports sending/receiving of MAD packets.
2856 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2858 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2862 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2863 * Management Datagrams.
2864 * @device: Device to check
2865 * @port_num: Port number to check
2867 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2868 * datagrams with their own versions. These OPA MADs share many but not all of
2869 * the characteristics of InfiniBand MADs.
2871 * OPA MADs differ in the following ways:
2873 * 1) MADs are variable size up to 2K
2874 * IBTA defined MADs remain fixed at 256 bytes
2875 * 2) OPA SMPs must carry valid PKeys
2876 * 3) OPA SMP packets are a different format
2878 * Return: true if the port supports OPA MAD packet formats.
2880 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2882 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2883 == RDMA_CORE_CAP_OPA_MAD;
2887 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2888 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2889 * @device: Device to check
2890 * @port_num: Port number to check
2892 * Each InfiniBand node is required to provide a Subnet Management Agent
2893 * that the subnet manager can access. Prior to the fabric being fully
2894 * configured by the subnet manager, the SMA is accessed via a well known
2895 * interface called the Subnet Management Interface (SMI). This interface
2896 * uses directed route packets to communicate with the SM to get around the
2897 * chicken and egg problem of the SM needing to know what's on the fabric
2898 * in order to configure the fabric, and needing to configure the fabric in
2899 * order to send packets to the devices on the fabric. These directed
2900 * route packets do not need the fabric fully configured in order to reach
2901 * their destination. The SMI is the only method allowed to send
2902 * directed route packets on an InfiniBand fabric.
2904 * Return: true if the port provides an SMI.
2906 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2908 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2912 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2913 * Communication Manager.
2914 * @device: Device to check
2915 * @port_num: Port number to check
2917 * The InfiniBand Communication Manager is one of many pre-defined General
2918 * Service Agents (GSA) that are accessed via the General Service
2919 * Interface (GSI). It's role is to facilitate establishment of connections
2920 * between nodes as well as other management related tasks for established
2923 * Return: true if the port supports an IB CM (this does not guarantee that
2924 * a CM is actually running however).
2926 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2928 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2932 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2933 * Communication Manager.
2934 * @device: Device to check
2935 * @port_num: Port number to check
2937 * Similar to above, but specific to iWARP connections which have a different
2938 * managment protocol than InfiniBand.
2940 * Return: true if the port supports an iWARP CM (this does not guarantee that
2941 * a CM is actually running however).
2943 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2945 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2949 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2950 * Subnet Administration.
2951 * @device: Device to check
2952 * @port_num: Port number to check
2954 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2955 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2956 * fabrics, devices should resolve routes to other hosts by contacting the
2957 * SA to query the proper route.
2959 * Return: true if the port should act as a client to the fabric Subnet
2960 * Administration interface. This does not imply that the SA service is
2963 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2965 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2969 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2971 * @device: Device to check
2972 * @port_num: Port number to check
2974 * InfiniBand multicast registration is more complex than normal IPv4 or
2975 * IPv6 multicast registration. Each Host Channel Adapter must register
2976 * with the Subnet Manager when it wishes to join a multicast group. It
2977 * should do so only once regardless of how many queue pairs it subscribes
2978 * to this group. And it should leave the group only after all queue pairs
2979 * attached to the group have been detached.
2981 * Return: true if the port must undertake the additional adminstrative
2982 * overhead of registering/unregistering with the SM and tracking of the
2983 * total number of queue pairs attached to the multicast group.
2985 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2987 return rdma_cap_ib_sa(device, port_num);
2991 * rdma_cap_af_ib - Check if the port of device has the capability
2992 * Native Infiniband Address.
2993 * @device: Device to check
2994 * @port_num: Port number to check
2996 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2997 * GID. RoCE uses a different mechanism, but still generates a GID via
2998 * a prescribed mechanism and port specific data.
3000 * Return: true if the port uses a GID address to identify devices on the
3003 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3005 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3009 * rdma_cap_eth_ah - Check if the port of device has the capability
3010 * Ethernet Address Handle.
3011 * @device: Device to check
3012 * @port_num: Port number to check
3014 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3015 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3016 * port. Normally, packet headers are generated by the sending host
3017 * adapter, but when sending connectionless datagrams, we must manually
3018 * inject the proper headers for the fabric we are communicating over.
3020 * Return: true if we are running as a RoCE port and must force the
3021 * addition of a Global Route Header built from our Ethernet Address
3022 * Handle into our header list for connectionless packets.
3024 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3026 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3030 * rdma_cap_opa_ah - Check if the port of device supports
3031 * OPA Address handles
3032 * @device: Device to check
3033 * @port_num: Port number to check
3035 * Return: true if we are running on an OPA device which supports
3036 * the extended OPA addressing.
3038 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3040 return (device->port_immutable[port_num].core_cap_flags &
3041 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3045 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3048 * @port_num: Port number
3050 * This MAD size includes the MAD headers and MAD payload. No other headers
3053 * Return the max MAD size required by the Port. Will return 0 if the port
3054 * does not support MADs
3056 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3058 return device->port_immutable[port_num].max_mad_size;
3062 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3063 * @device: Device to check
3064 * @port_num: Port number to check
3066 * RoCE GID table mechanism manages the various GIDs for a device.
3068 * NOTE: if allocating the port's GID table has failed, this call will still
3069 * return true, but any RoCE GID table API will fail.
3071 * Return: true if the port uses RoCE GID table mechanism in order to manage
3074 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3077 return rdma_protocol_roce(device, port_num) &&
3078 device->add_gid && device->del_gid;
3082 * Check if the device supports READ W/ INVALIDATE.
3084 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3087 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3088 * has support for it yet.
3090 return rdma_protocol_iwarp(dev, port_num);
3093 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3095 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3096 struct ifla_vf_info *info);
3097 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3098 struct ifla_vf_stats *stats);
3099 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3102 int ib_query_pkey(struct ib_device *device,
3103 u8 port_num, u16 index, u16 *pkey);
3105 int ib_modify_device(struct ib_device *device,
3106 int device_modify_mask,
3107 struct ib_device_modify *device_modify);
3109 int ib_modify_port(struct ib_device *device,
3110 u8 port_num, int port_modify_mask,
3111 struct ib_port_modify *port_modify);
3113 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3114 u8 *port_num, u16 *index);
3116 int ib_find_pkey(struct ib_device *device,
3117 u8 port_num, u16 pkey, u16 *index);
3121 * Create a memory registration for all memory in the system and place
3122 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3123 * ULPs to avoid the overhead of dynamic MRs.
3125 * This flag is generally considered unsafe and must only be used in
3126 * extremly trusted environments. Every use of it will log a warning
3127 * in the kernel log.
3129 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3132 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3133 const char *caller);
3134 #define ib_alloc_pd(device, flags) \
3135 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3136 void ib_dealloc_pd(struct ib_pd *pd);
3139 * rdma_create_ah - Creates an address handle for the given address vector.
3140 * @pd: The protection domain associated with the address handle.
3141 * @ah_attr: The attributes of the address vector.
3143 * The address handle is used to reference a local or global destination
3144 * in all UD QP post sends.
3146 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3149 * rdma_create_user_ah - Creates an address handle for the given address vector.
3150 * It resolves destination mac address for ah attribute of RoCE type.
3151 * @pd: The protection domain associated with the address handle.
3152 * @ah_attr: The attributes of the address vector.
3153 * @udata: pointer to user's input output buffer information need by
3156 * It returns 0 on success and returns appropriate error code on error.
3157 * The address handle is used to reference a local or global destination
3158 * in all UD QP post sends.
3160 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3161 struct rdma_ah_attr *ah_attr,
3162 struct ib_udata *udata);
3164 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3166 * @hdr: the L3 header to parse
3167 * @net_type: type of header to parse
3168 * @sgid: place to store source gid
3169 * @dgid: place to store destination gid
3171 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3172 enum rdma_network_type net_type,
3173 union ib_gid *sgid, union ib_gid *dgid);
3176 * ib_get_rdma_header_version - Get the header version
3177 * @hdr: the L3 header to parse
3179 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3182 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3184 * @device: Device on which the received message arrived.
3185 * @port_num: Port on which the received message arrived.
3186 * @wc: Work completion associated with the received message.
3187 * @grh: References the received global route header. This parameter is
3188 * ignored unless the work completion indicates that the GRH is valid.
3189 * @ah_attr: Returned attributes that can be used when creating an address
3190 * handle for replying to the message.
3191 * When ib_init_ah_attr_from_wc() returns success,
3192 * (a) for IB link layer it optionally contains a reference to SGID attribute
3193 * when GRH is present for IB link layer.
3194 * (b) for RoCE link layer it contains a reference to SGID attribute.
3195 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3196 * attributes which are initialized using ib_init_ah_attr_from_wc().
3199 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3200 const struct ib_wc *wc, const struct ib_grh *grh,
3201 struct rdma_ah_attr *ah_attr);
3204 * ib_create_ah_from_wc - Creates an address handle associated with the
3205 * sender of the specified work completion.
3206 * @pd: The protection domain associated with the address handle.
3207 * @wc: Work completion information associated with a received message.
3208 * @grh: References the received global route header. This parameter is
3209 * ignored unless the work completion indicates that the GRH is valid.
3210 * @port_num: The outbound port number to associate with the address.
3212 * The address handle is used to reference a local or global destination
3213 * in all UD QP post sends.
3215 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3216 const struct ib_grh *grh, u8 port_num);
3219 * rdma_modify_ah - Modifies the address vector associated with an address
3221 * @ah: The address handle to modify.
3222 * @ah_attr: The new address vector attributes to associate with the
3225 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3228 * rdma_query_ah - Queries the address vector associated with an address
3230 * @ah: The address handle to query.
3231 * @ah_attr: The address vector attributes associated with the address
3234 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3237 * rdma_destroy_ah - Destroys an address handle.
3238 * @ah: The address handle to destroy.
3240 int rdma_destroy_ah(struct ib_ah *ah);
3243 * ib_create_srq - Creates a SRQ associated with the specified protection
3245 * @pd: The protection domain associated with the SRQ.
3246 * @srq_init_attr: A list of initial attributes required to create the
3247 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3248 * the actual capabilities of the created SRQ.
3250 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3251 * requested size of the SRQ, and set to the actual values allocated
3252 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3253 * will always be at least as large as the requested values.
3255 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3256 struct ib_srq_init_attr *srq_init_attr);
3259 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3260 * @srq: The SRQ to modify.
3261 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3262 * the current values of selected SRQ attributes are returned.
3263 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3264 * are being modified.
3266 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3267 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3268 * the number of receives queued drops below the limit.
3270 int ib_modify_srq(struct ib_srq *srq,
3271 struct ib_srq_attr *srq_attr,
3272 enum ib_srq_attr_mask srq_attr_mask);
3275 * ib_query_srq - Returns the attribute list and current values for the
3277 * @srq: The SRQ to query.
3278 * @srq_attr: The attributes of the specified SRQ.
3280 int ib_query_srq(struct ib_srq *srq,
3281 struct ib_srq_attr *srq_attr);
3284 * ib_destroy_srq - Destroys the specified SRQ.
3285 * @srq: The SRQ to destroy.
3287 int ib_destroy_srq(struct ib_srq *srq);
3290 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3291 * @srq: The SRQ to post the work request on.
3292 * @recv_wr: A list of work requests to post on the receive queue.
3293 * @bad_recv_wr: On an immediate failure, this parameter will reference
3294 * the work request that failed to be posted on the QP.
3296 static inline int ib_post_srq_recv(struct ib_srq *srq,
3297 const struct ib_recv_wr *recv_wr,
3298 const struct ib_recv_wr **bad_recv_wr)
3300 const struct ib_recv_wr *dummy;
3302 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr ? : &dummy);
3306 * ib_create_qp - Creates a QP associated with the specified protection
3308 * @pd: The protection domain associated with the QP.
3309 * @qp_init_attr: A list of initial attributes required to create the
3310 * QP. If QP creation succeeds, then the attributes are updated to
3311 * the actual capabilities of the created QP.
3313 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3314 struct ib_qp_init_attr *qp_init_attr);
3317 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3318 * @qp: The QP to modify.
3319 * @attr: On input, specifies the QP attributes to modify. On output,
3320 * the current values of selected QP attributes are returned.
3321 * @attr_mask: A bit-mask used to specify which attributes of the QP
3322 * are being modified.
3323 * @udata: pointer to user's input output buffer information
3324 * are being modified.
3325 * It returns 0 on success and returns appropriate error code on error.
3327 int ib_modify_qp_with_udata(struct ib_qp *qp,
3328 struct ib_qp_attr *attr,
3330 struct ib_udata *udata);
3333 * ib_modify_qp - Modifies the attributes for the specified QP and then
3334 * transitions the QP to the given state.
3335 * @qp: The QP to modify.
3336 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3337 * the current values of selected QP attributes are returned.
3338 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3339 * are being modified.
3341 int ib_modify_qp(struct ib_qp *qp,
3342 struct ib_qp_attr *qp_attr,
3346 * ib_query_qp - Returns the attribute list and current values for the
3348 * @qp: The QP to query.
3349 * @qp_attr: The attributes of the specified QP.
3350 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3351 * @qp_init_attr: Additional attributes of the selected QP.
3353 * The qp_attr_mask may be used to limit the query to gathering only the
3354 * selected attributes.
3356 int ib_query_qp(struct ib_qp *qp,
3357 struct ib_qp_attr *qp_attr,
3359 struct ib_qp_init_attr *qp_init_attr);
3362 * ib_destroy_qp - Destroys the specified QP.
3363 * @qp: The QP to destroy.
3365 int ib_destroy_qp(struct ib_qp *qp);
3368 * ib_open_qp - Obtain a reference to an existing sharable QP.
3369 * @xrcd - XRC domain
3370 * @qp_open_attr: Attributes identifying the QP to open.
3372 * Returns a reference to a sharable QP.
3374 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3375 struct ib_qp_open_attr *qp_open_attr);
3378 * ib_close_qp - Release an external reference to a QP.
3379 * @qp: The QP handle to release
3381 * The opened QP handle is released by the caller. The underlying
3382 * shared QP is not destroyed until all internal references are released.
3384 int ib_close_qp(struct ib_qp *qp);
3387 * ib_post_send - Posts a list of work requests to the send queue of
3389 * @qp: The QP to post the work request on.
3390 * @send_wr: A list of work requests to post on the send queue.
3391 * @bad_send_wr: On an immediate failure, this parameter will reference
3392 * the work request that failed to be posted on the QP.
3394 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3395 * error is returned, the QP state shall not be affected,
3396 * ib_post_send() will return an immediate error after queueing any
3397 * earlier work requests in the list.
3399 static inline int ib_post_send(struct ib_qp *qp,
3400 const struct ib_send_wr *send_wr,
3401 const struct ib_send_wr **bad_send_wr)
3403 const struct ib_send_wr *dummy;
3405 return qp->device->post_send(qp, send_wr, bad_send_wr ? : &dummy);
3409 * ib_post_recv - Posts a list of work requests to the receive queue of
3411 * @qp: The QP to post the work request on.
3412 * @recv_wr: A list of work requests to post on the receive queue.
3413 * @bad_recv_wr: On an immediate failure, this parameter will reference
3414 * the work request that failed to be posted on the QP.
3416 static inline int ib_post_recv(struct ib_qp *qp,
3417 const struct ib_recv_wr *recv_wr,
3418 const struct ib_recv_wr **bad_recv_wr)
3420 const struct ib_recv_wr *dummy;
3422 return qp->device->post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3425 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3426 int nr_cqe, int comp_vector,
3427 enum ib_poll_context poll_ctx, const char *caller);
3428 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3429 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3431 void ib_free_cq(struct ib_cq *cq);
3432 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3435 * ib_create_cq - Creates a CQ on the specified device.
3436 * @device: The device on which to create the CQ.
3437 * @comp_handler: A user-specified callback that is invoked when a
3438 * completion event occurs on the CQ.
3439 * @event_handler: A user-specified callback that is invoked when an
3440 * asynchronous event not associated with a completion occurs on the CQ.
3441 * @cq_context: Context associated with the CQ returned to the user via
3442 * the associated completion and event handlers.
3443 * @cq_attr: The attributes the CQ should be created upon.
3445 * Users can examine the cq structure to determine the actual CQ size.
3447 struct ib_cq *__ib_create_cq(struct ib_device *device,
3448 ib_comp_handler comp_handler,
3449 void (*event_handler)(struct ib_event *, void *),
3451 const struct ib_cq_init_attr *cq_attr,
3452 const char *caller);
3453 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3454 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3457 * ib_resize_cq - Modifies the capacity of the CQ.
3458 * @cq: The CQ to resize.
3459 * @cqe: The minimum size of the CQ.
3461 * Users can examine the cq structure to determine the actual CQ size.
3463 int ib_resize_cq(struct ib_cq *cq, int cqe);
3466 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3467 * @cq: The CQ to modify.
3468 * @cq_count: number of CQEs that will trigger an event
3469 * @cq_period: max period of time in usec before triggering an event
3472 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3475 * ib_destroy_cq - Destroys the specified CQ.
3476 * @cq: The CQ to destroy.
3478 int ib_destroy_cq(struct ib_cq *cq);
3481 * ib_poll_cq - poll a CQ for completion(s)
3482 * @cq:the CQ being polled
3483 * @num_entries:maximum number of completions to return
3484 * @wc:array of at least @num_entries &struct ib_wc where completions
3487 * Poll a CQ for (possibly multiple) completions. If the return value
3488 * is < 0, an error occurred. If the return value is >= 0, it is the
3489 * number of completions returned. If the return value is
3490 * non-negative and < num_entries, then the CQ was emptied.
3492 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3495 return cq->device->poll_cq(cq, num_entries, wc);
3499 * ib_req_notify_cq - Request completion notification on a CQ.
3500 * @cq: The CQ to generate an event for.
3502 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3503 * to request an event on the next solicited event or next work
3504 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3505 * may also be |ed in to request a hint about missed events, as
3509 * < 0 means an error occurred while requesting notification
3510 * == 0 means notification was requested successfully, and if
3511 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3512 * were missed and it is safe to wait for another event. In
3513 * this case is it guaranteed that any work completions added
3514 * to the CQ since the last CQ poll will trigger a completion
3515 * notification event.
3516 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3517 * in. It means that the consumer must poll the CQ again to
3518 * make sure it is empty to avoid missing an event because of a
3519 * race between requesting notification and an entry being
3520 * added to the CQ. This return value means it is possible
3521 * (but not guaranteed) that a work completion has been added
3522 * to the CQ since the last poll without triggering a
3523 * completion notification event.
3525 static inline int ib_req_notify_cq(struct ib_cq *cq,
3526 enum ib_cq_notify_flags flags)
3528 return cq->device->req_notify_cq(cq, flags);
3532 * ib_req_ncomp_notif - Request completion notification when there are
3533 * at least the specified number of unreaped completions on the CQ.
3534 * @cq: The CQ to generate an event for.
3535 * @wc_cnt: The number of unreaped completions that should be on the
3536 * CQ before an event is generated.
3538 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3540 return cq->device->req_ncomp_notif ?
3541 cq->device->req_ncomp_notif(cq, wc_cnt) :
3546 * ib_dma_mapping_error - check a DMA addr for error
3547 * @dev: The device for which the dma_addr was created
3548 * @dma_addr: The DMA address to check
3550 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3552 return dma_mapping_error(dev->dma_device, dma_addr);
3556 * ib_dma_map_single - Map a kernel virtual address to DMA address
3557 * @dev: The device for which the dma_addr is to be created
3558 * @cpu_addr: The kernel virtual address
3559 * @size: The size of the region in bytes
3560 * @direction: The direction of the DMA
3562 static inline u64 ib_dma_map_single(struct ib_device *dev,
3563 void *cpu_addr, size_t size,
3564 enum dma_data_direction direction)
3566 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3570 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3571 * @dev: The device for which the DMA address was created
3572 * @addr: The DMA address
3573 * @size: The size of the region in bytes
3574 * @direction: The direction of the DMA
3576 static inline void ib_dma_unmap_single(struct ib_device *dev,
3577 u64 addr, size_t size,
3578 enum dma_data_direction direction)
3580 dma_unmap_single(dev->dma_device, addr, size, direction);
3584 * ib_dma_map_page - Map a physical page to DMA address
3585 * @dev: The device for which the dma_addr is to be created
3586 * @page: The page to be mapped
3587 * @offset: The offset within the page
3588 * @size: The size of the region in bytes
3589 * @direction: The direction of the DMA
3591 static inline u64 ib_dma_map_page(struct ib_device *dev,
3593 unsigned long offset,
3595 enum dma_data_direction direction)
3597 return dma_map_page(dev->dma_device, page, offset, size, direction);
3601 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3602 * @dev: The device for which the DMA address was created
3603 * @addr: The DMA address
3604 * @size: The size of the region in bytes
3605 * @direction: The direction of the DMA
3607 static inline void ib_dma_unmap_page(struct ib_device *dev,
3608 u64 addr, size_t size,
3609 enum dma_data_direction direction)
3611 dma_unmap_page(dev->dma_device, addr, size, direction);
3615 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3616 * @dev: The device for which the DMA addresses are to be created
3617 * @sg: The array of scatter/gather entries
3618 * @nents: The number of scatter/gather entries
3619 * @direction: The direction of the DMA
3621 static inline int ib_dma_map_sg(struct ib_device *dev,
3622 struct scatterlist *sg, int nents,
3623 enum dma_data_direction direction)
3625 return dma_map_sg(dev->dma_device, sg, nents, direction);
3629 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3630 * @dev: The device for which the DMA addresses were created
3631 * @sg: The array of scatter/gather entries
3632 * @nents: The number of scatter/gather entries
3633 * @direction: The direction of the DMA
3635 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3636 struct scatterlist *sg, int nents,
3637 enum dma_data_direction direction)
3639 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3642 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3643 struct scatterlist *sg, int nents,
3644 enum dma_data_direction direction,
3645 unsigned long dma_attrs)
3647 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3651 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3652 struct scatterlist *sg, int nents,
3653 enum dma_data_direction direction,
3654 unsigned long dma_attrs)
3656 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3659 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3660 * @dev: The device for which the DMA addresses were created
3661 * @sg: The scatter/gather entry
3663 * Note: this function is obsolete. To do: change all occurrences of
3664 * ib_sg_dma_address() into sg_dma_address().
3666 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3667 struct scatterlist *sg)
3669 return sg_dma_address(sg);
3673 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3674 * @dev: The device for which the DMA addresses were created
3675 * @sg: The scatter/gather entry
3677 * Note: this function is obsolete. To do: change all occurrences of
3678 * ib_sg_dma_len() into sg_dma_len().
3680 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3681 struct scatterlist *sg)
3683 return sg_dma_len(sg);
3687 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3688 * @dev: The device for which the DMA address was created
3689 * @addr: The DMA address
3690 * @size: The size of the region in bytes
3691 * @dir: The direction of the DMA
3693 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3696 enum dma_data_direction dir)
3698 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3702 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3703 * @dev: The device for which the DMA address was created
3704 * @addr: The DMA address
3705 * @size: The size of the region in bytes
3706 * @dir: The direction of the DMA
3708 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3711 enum dma_data_direction dir)
3713 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3717 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3718 * @dev: The device for which the DMA address is requested
3719 * @size: The size of the region to allocate in bytes
3720 * @dma_handle: A pointer for returning the DMA address of the region
3721 * @flag: memory allocator flags
3723 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3725 dma_addr_t *dma_handle,
3728 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3732 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3733 * @dev: The device for which the DMA addresses were allocated
3734 * @size: The size of the region
3735 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3736 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3738 static inline void ib_dma_free_coherent(struct ib_device *dev,
3739 size_t size, void *cpu_addr,
3740 dma_addr_t dma_handle)
3742 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3746 * ib_dereg_mr - Deregisters a memory region and removes it from the
3747 * HCA translation table.
3748 * @mr: The memory region to deregister.
3750 * This function can fail, if the memory region has memory windows bound to it.
3752 int ib_dereg_mr(struct ib_mr *mr);
3754 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3755 enum ib_mr_type mr_type,
3759 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3761 * @mr - struct ib_mr pointer to be updated.
3762 * @newkey - new key to be used.
3764 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3766 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3767 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3771 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3772 * for calculating a new rkey for type 2 memory windows.
3773 * @rkey - the rkey to increment.
3775 static inline u32 ib_inc_rkey(u32 rkey)
3777 const u32 mask = 0x000000ff;
3778 return ((rkey + 1) & mask) | (rkey & ~mask);
3782 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3783 * @pd: The protection domain associated with the unmapped region.
3784 * @mr_access_flags: Specifies the memory access rights.
3785 * @fmr_attr: Attributes of the unmapped region.
3787 * A fast memory region must be mapped before it can be used as part of
3790 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3791 int mr_access_flags,
3792 struct ib_fmr_attr *fmr_attr);
3795 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3796 * @fmr: The fast memory region to associate with the pages.
3797 * @page_list: An array of physical pages to map to the fast memory region.
3798 * @list_len: The number of pages in page_list.
3799 * @iova: The I/O virtual address to use with the mapped region.
3801 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3802 u64 *page_list, int list_len,
3805 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3809 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3810 * @fmr_list: A linked list of fast memory regions to unmap.
3812 int ib_unmap_fmr(struct list_head *fmr_list);
3815 * ib_dealloc_fmr - Deallocates a fast memory region.
3816 * @fmr: The fast memory region to deallocate.
3818 int ib_dealloc_fmr(struct ib_fmr *fmr);
3821 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3822 * @qp: QP to attach to the multicast group. The QP must be type
3824 * @gid: Multicast group GID.
3825 * @lid: Multicast group LID in host byte order.
3827 * In order to send and receive multicast packets, subnet
3828 * administration must have created the multicast group and configured
3829 * the fabric appropriately. The port associated with the specified
3830 * QP must also be a member of the multicast group.
3832 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3835 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3836 * @qp: QP to detach from the multicast group.
3837 * @gid: Multicast group GID.
3838 * @lid: Multicast group LID in host byte order.
3840 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3843 * ib_alloc_xrcd - Allocates an XRC domain.
3844 * @device: The device on which to allocate the XRC domain.
3845 * @caller: Module name for kernel consumers
3847 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3848 #define ib_alloc_xrcd(device) \
3849 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3852 * ib_dealloc_xrcd - Deallocates an XRC domain.
3853 * @xrcd: The XRC domain to deallocate.
3855 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3857 static inline int ib_check_mr_access(int flags)
3860 * Local write permission is required if remote write or
3861 * remote atomic permission is also requested.
3863 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3864 !(flags & IB_ACCESS_LOCAL_WRITE))
3867 if (flags & ~IB_ACCESS_SUPPORTED)
3873 static inline bool ib_access_writable(int access_flags)
3876 * We have writable memory backing the MR if any of the following
3877 * access flags are set. "Local write" and "remote write" obviously
3878 * require write access. "Remote atomic" can do things like fetch and
3879 * add, which will modify memory, and "MW bind" can change permissions
3880 * by binding a window.
3882 return access_flags &
3883 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3884 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3888 * ib_check_mr_status: lightweight check of MR status.
3889 * This routine may provide status checks on a selected
3890 * ib_mr. first use is for signature status check.
3892 * @mr: A memory region.
3893 * @check_mask: Bitmask of which checks to perform from
3894 * ib_mr_status_check enumeration.
3895 * @mr_status: The container of relevant status checks.
3896 * failed checks will be indicated in the status bitmask
3897 * and the relevant info shall be in the error item.
3899 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3900 struct ib_mr_status *mr_status);
3902 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3903 u16 pkey, const union ib_gid *gid,
3904 const struct sockaddr *addr);
3905 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3906 struct ib_wq_init_attr *init_attr);
3907 int ib_destroy_wq(struct ib_wq *wq);
3908 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3910 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3911 struct ib_rwq_ind_table_init_attr*
3912 wq_ind_table_init_attr);
3913 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3915 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3916 unsigned int *sg_offset, unsigned int page_size);
3919 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3920 unsigned int *sg_offset, unsigned int page_size)
3924 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3930 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3931 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3933 void ib_drain_rq(struct ib_qp *qp);
3934 void ib_drain_sq(struct ib_qp *qp);
3935 void ib_drain_qp(struct ib_qp *qp);
3937 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3939 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3941 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3942 return attr->roce.dmac;
3946 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3948 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3949 attr->ib.dlid = (u16)dlid;
3950 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3951 attr->opa.dlid = dlid;
3954 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3956 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3957 return attr->ib.dlid;
3958 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3959 return attr->opa.dlid;
3963 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3968 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3973 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3976 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3977 attr->ib.src_path_bits = src_path_bits;
3978 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3979 attr->opa.src_path_bits = src_path_bits;
3982 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3984 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3985 return attr->ib.src_path_bits;
3986 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3987 return attr->opa.src_path_bits;
3991 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3994 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3995 attr->opa.make_grd = make_grd;
3998 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4000 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4001 return attr->opa.make_grd;
4005 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4007 attr->port_num = port_num;
4010 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4012 return attr->port_num;
4015 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4018 attr->static_rate = static_rate;
4021 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4023 return attr->static_rate;
4026 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4027 enum ib_ah_flags flag)
4029 attr->ah_flags = flag;
4032 static inline enum ib_ah_flags
4033 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4035 return attr->ah_flags;
4038 static inline const struct ib_global_route
4039 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4044 /*To retrieve and modify the grh */
4045 static inline struct ib_global_route
4046 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4051 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4053 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4055 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4058 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4061 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4063 grh->dgid.global.subnet_prefix = prefix;
4066 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4069 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4071 grh->dgid.global.interface_id = if_id;
4074 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4075 union ib_gid *dgid, u32 flow_label,
4076 u8 sgid_index, u8 hop_limit,
4079 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4081 attr->ah_flags = IB_AH_GRH;
4084 grh->flow_label = flow_label;
4085 grh->sgid_index = sgid_index;
4086 grh->hop_limit = hop_limit;
4087 grh->traffic_class = traffic_class;
4088 grh->sgid_attr = NULL;
4091 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4092 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4093 u32 flow_label, u8 hop_limit, u8 traffic_class,
4094 const struct ib_gid_attr *sgid_attr);
4095 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4096 const struct rdma_ah_attr *src);
4097 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4098 const struct rdma_ah_attr *new);
4099 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4102 * rdma_ah_find_type - Return address handle type.
4104 * @dev: Device to be checked
4105 * @port_num: Port number
4107 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4110 if (rdma_protocol_roce(dev, port_num))
4111 return RDMA_AH_ATTR_TYPE_ROCE;
4112 if (rdma_protocol_ib(dev, port_num)) {
4113 if (rdma_cap_opa_ah(dev, port_num))
4114 return RDMA_AH_ATTR_TYPE_OPA;
4115 return RDMA_AH_ATTR_TYPE_IB;
4118 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4122 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4123 * In the current implementation the only way to get
4124 * get the 32bit lid is from other sources for OPA.
4125 * For IB, lids will always be 16bits so cast the
4126 * value accordingly.
4130 static inline u16 ib_lid_cpu16(u32 lid)
4132 WARN_ON_ONCE(lid & 0xFFFF0000);
4137 * ib_lid_be16 - Return lid in 16bit BE encoding.
4141 static inline __be16 ib_lid_be16(u32 lid)
4143 WARN_ON_ONCE(lid & 0xFFFF0000);
4144 return cpu_to_be16((u16)lid);
4148 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4150 * @device: the rdma device
4151 * @comp_vector: index of completion vector
4153 * Returns NULL on failure, otherwise a corresponding cpu map of the
4154 * completion vector (returns all-cpus map if the device driver doesn't
4155 * implement get_vector_affinity).
4157 static inline const struct cpumask *
4158 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4160 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4161 !device->get_vector_affinity)
4164 return device->get_vector_affinity(device, comp_vector);
4168 static inline void ib_set_flow(struct ib_uobject *uobj, struct ib_flow *ibflow,
4169 struct ib_qp *qp, struct ib_device *device)
4171 uobj->object = ibflow;
4172 ibflow->uobject = uobj;
4175 atomic_inc(&qp->usecnt);
4179 ibflow->device = device;
4183 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4184 * and add their gids, as needed, to the relevant RoCE devices.
4186 * @device: the rdma device
4188 void rdma_roce_rescan_device(struct ib_device *ibdev);
4190 struct ib_ucontext *ib_uverbs_get_ucontext(struct ib_uverbs_file *ufile);
4192 int uverbs_destroy_def_handler(struct ib_uverbs_file *file,
4193 struct uverbs_attr_bundle *attrs);
4194 #endif /* IB_VERBS_H */