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 on RC transport */
295 IB_TM_CAP_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 /* Consumer's event_handler callback must not block */
1142 void (*event_handler)(struct ib_event *, void *);
1145 struct ib_cq *send_cq;
1146 struct ib_cq *recv_cq;
1148 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1149 struct ib_qp_cap cap;
1150 enum ib_sig_type sq_sig_type;
1151 enum ib_qp_type qp_type;
1152 enum ib_qp_create_flags create_flags;
1155 * Only needed for special QP types, or when using the RW API.
1158 struct ib_rwq_ind_table *rwq_ind_tbl;
1162 struct ib_qp_open_attr {
1163 void (*event_handler)(struct ib_event *, void *);
1166 enum ib_qp_type qp_type;
1169 enum ib_rnr_timeout {
1170 IB_RNR_TIMER_655_36 = 0,
1171 IB_RNR_TIMER_000_01 = 1,
1172 IB_RNR_TIMER_000_02 = 2,
1173 IB_RNR_TIMER_000_03 = 3,
1174 IB_RNR_TIMER_000_04 = 4,
1175 IB_RNR_TIMER_000_06 = 5,
1176 IB_RNR_TIMER_000_08 = 6,
1177 IB_RNR_TIMER_000_12 = 7,
1178 IB_RNR_TIMER_000_16 = 8,
1179 IB_RNR_TIMER_000_24 = 9,
1180 IB_RNR_TIMER_000_32 = 10,
1181 IB_RNR_TIMER_000_48 = 11,
1182 IB_RNR_TIMER_000_64 = 12,
1183 IB_RNR_TIMER_000_96 = 13,
1184 IB_RNR_TIMER_001_28 = 14,
1185 IB_RNR_TIMER_001_92 = 15,
1186 IB_RNR_TIMER_002_56 = 16,
1187 IB_RNR_TIMER_003_84 = 17,
1188 IB_RNR_TIMER_005_12 = 18,
1189 IB_RNR_TIMER_007_68 = 19,
1190 IB_RNR_TIMER_010_24 = 20,
1191 IB_RNR_TIMER_015_36 = 21,
1192 IB_RNR_TIMER_020_48 = 22,
1193 IB_RNR_TIMER_030_72 = 23,
1194 IB_RNR_TIMER_040_96 = 24,
1195 IB_RNR_TIMER_061_44 = 25,
1196 IB_RNR_TIMER_081_92 = 26,
1197 IB_RNR_TIMER_122_88 = 27,
1198 IB_RNR_TIMER_163_84 = 28,
1199 IB_RNR_TIMER_245_76 = 29,
1200 IB_RNR_TIMER_327_68 = 30,
1201 IB_RNR_TIMER_491_52 = 31
1204 enum ib_qp_attr_mask {
1206 IB_QP_CUR_STATE = (1<<1),
1207 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1208 IB_QP_ACCESS_FLAGS = (1<<3),
1209 IB_QP_PKEY_INDEX = (1<<4),
1210 IB_QP_PORT = (1<<5),
1211 IB_QP_QKEY = (1<<6),
1213 IB_QP_PATH_MTU = (1<<8),
1214 IB_QP_TIMEOUT = (1<<9),
1215 IB_QP_RETRY_CNT = (1<<10),
1216 IB_QP_RNR_RETRY = (1<<11),
1217 IB_QP_RQ_PSN = (1<<12),
1218 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1219 IB_QP_ALT_PATH = (1<<14),
1220 IB_QP_MIN_RNR_TIMER = (1<<15),
1221 IB_QP_SQ_PSN = (1<<16),
1222 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1223 IB_QP_PATH_MIG_STATE = (1<<18),
1224 IB_QP_CAP = (1<<19),
1225 IB_QP_DEST_QPN = (1<<20),
1226 IB_QP_RESERVED1 = (1<<21),
1227 IB_QP_RESERVED2 = (1<<22),
1228 IB_QP_RESERVED3 = (1<<23),
1229 IB_QP_RESERVED4 = (1<<24),
1230 IB_QP_RATE_LIMIT = (1<<25),
1255 enum ib_qp_state qp_state;
1256 enum ib_qp_state cur_qp_state;
1257 enum ib_mtu path_mtu;
1258 enum ib_mig_state path_mig_state;
1263 int qp_access_flags;
1264 struct ib_qp_cap cap;
1265 struct rdma_ah_attr ah_attr;
1266 struct rdma_ah_attr alt_ah_attr;
1269 u8 en_sqd_async_notify;
1272 u8 max_dest_rd_atomic;
1284 /* These are shared with userspace */
1285 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1286 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1287 IB_WR_SEND = IB_UVERBS_WR_SEND,
1288 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1289 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1290 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1291 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1292 IB_WR_LSO = IB_UVERBS_WR_TSO,
1293 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1294 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1295 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1296 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1297 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1298 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1299 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1301 /* These are kernel only and can not be issued by userspace */
1302 IB_WR_REG_MR = 0x20,
1305 /* reserve values for low level drivers' internal use.
1306 * These values will not be used at all in the ib core layer.
1308 IB_WR_RESERVED1 = 0xf0,
1320 enum ib_send_flags {
1322 IB_SEND_SIGNALED = (1<<1),
1323 IB_SEND_SOLICITED = (1<<2),
1324 IB_SEND_INLINE = (1<<3),
1325 IB_SEND_IP_CSUM = (1<<4),
1327 /* reserve bits 26-31 for low level drivers' internal use */
1328 IB_SEND_RESERVED_START = (1 << 26),
1329 IB_SEND_RESERVED_END = (1 << 31),
1339 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1343 struct ib_send_wr *next;
1346 struct ib_cqe *wr_cqe;
1348 struct ib_sge *sg_list;
1350 enum ib_wr_opcode opcode;
1354 u32 invalidate_rkey;
1359 struct ib_send_wr wr;
1364 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1366 return container_of(wr, struct ib_rdma_wr, wr);
1369 struct ib_atomic_wr {
1370 struct ib_send_wr wr;
1374 u64 compare_add_mask;
1379 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1381 return container_of(wr, struct ib_atomic_wr, wr);
1385 struct ib_send_wr wr;
1392 u16 pkey_index; /* valid for GSI only */
1393 u8 port_num; /* valid for DR SMPs on switch only */
1396 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1398 return container_of(wr, struct ib_ud_wr, wr);
1402 struct ib_send_wr wr;
1408 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1410 return container_of(wr, struct ib_reg_wr, wr);
1413 struct ib_sig_handover_wr {
1414 struct ib_send_wr wr;
1415 struct ib_sig_attrs *sig_attrs;
1416 struct ib_mr *sig_mr;
1418 struct ib_sge *prot;
1421 static inline const struct ib_sig_handover_wr *
1422 sig_handover_wr(const struct ib_send_wr *wr)
1424 return container_of(wr, struct ib_sig_handover_wr, wr);
1428 struct ib_recv_wr *next;
1431 struct ib_cqe *wr_cqe;
1433 struct ib_sge *sg_list;
1437 enum ib_access_flags {
1438 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1439 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1440 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1441 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1442 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1443 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1444 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1445 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1447 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1451 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1452 * are hidden here instead of a uapi header!
1454 enum ib_mr_rereg_flags {
1455 IB_MR_REREG_TRANS = 1,
1456 IB_MR_REREG_PD = (1<<1),
1457 IB_MR_REREG_ACCESS = (1<<2),
1458 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1461 struct ib_fmr_attr {
1469 enum rdma_remove_reason {
1471 * Userspace requested uobject deletion or initial try
1472 * to remove uobject via cleanup. Call could fail
1474 RDMA_REMOVE_DESTROY,
1475 /* Context deletion. This call should delete the actual object itself */
1477 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1478 RDMA_REMOVE_DRIVER_REMOVE,
1479 /* uobj is being cleaned-up before being committed */
1483 struct ib_rdmacg_object {
1484 #ifdef CONFIG_CGROUP_RDMA
1485 struct rdma_cgroup *cg; /* owner rdma cgroup */
1489 struct ib_ucontext {
1490 struct ib_device *device;
1491 struct ib_uverbs_file *ufile;
1493 * 'closing' can be read by the driver only during a destroy callback,
1494 * it is set when we are closing the file descriptor and indicates
1495 * that mm_sem may be locked.
1499 bool cleanup_retryable;
1502 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1503 struct rb_root_cached umem_tree;
1505 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1506 * mmu notifiers registration.
1508 struct rw_semaphore umem_rwsem;
1509 void (*invalidate_range)(struct ib_umem *umem,
1510 unsigned long start, unsigned long end);
1512 struct mmu_notifier mn;
1513 atomic_t notifier_count;
1514 /* A list of umems that don't have private mmu notifier counters yet. */
1515 struct list_head no_private_counters;
1519 struct ib_rdmacg_object cg_obj;
1523 u64 user_handle; /* handle given to us by userspace */
1524 /* ufile & ucontext owning this object */
1525 struct ib_uverbs_file *ufile;
1526 /* FIXME, save memory: ufile->context == context */
1527 struct ib_ucontext *context; /* associated user context */
1528 void *object; /* containing object */
1529 struct list_head list; /* link to context's list */
1530 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1531 int id; /* index into kernel idr */
1533 atomic_t usecnt; /* protects exclusive access */
1534 struct rcu_head rcu; /* kfree_rcu() overhead */
1536 const struct uverbs_api_object *uapi_object;
1540 const void __user *inbuf;
1541 void __user *outbuf;
1549 struct ib_device *device;
1550 struct ib_uobject *uobject;
1551 atomic_t usecnt; /* count all resources */
1553 u32 unsafe_global_rkey;
1556 * Implementation details of the RDMA core, don't use in drivers:
1558 struct ib_mr *__internal_mr;
1559 struct rdma_restrack_entry res;
1563 struct ib_device *device;
1564 atomic_t usecnt; /* count all exposed resources */
1565 struct inode *inode;
1567 struct mutex tgt_qp_mutex;
1568 struct list_head tgt_qp_list;
1572 struct ib_device *device;
1574 struct ib_uobject *uobject;
1575 const struct ib_gid_attr *sgid_attr;
1576 enum rdma_ah_attr_type type;
1579 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1581 enum ib_poll_context {
1582 IB_POLL_DIRECT, /* caller context, no hw completions */
1583 IB_POLL_SOFTIRQ, /* poll from softirq context */
1584 IB_POLL_WORKQUEUE, /* poll from workqueue */
1585 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1589 struct ib_device *device;
1590 struct ib_uobject *uobject;
1591 ib_comp_handler comp_handler;
1592 void (*event_handler)(struct ib_event *, void *);
1595 atomic_t usecnt; /* count number of work queues */
1596 enum ib_poll_context poll_ctx;
1599 struct irq_poll iop;
1600 struct work_struct work;
1602 struct workqueue_struct *comp_wq;
1604 * Implementation details of the RDMA core, don't use in drivers:
1606 struct rdma_restrack_entry res;
1610 struct ib_device *device;
1612 struct ib_uobject *uobject;
1613 void (*event_handler)(struct ib_event *, void *);
1615 enum ib_srq_type srq_type;
1622 struct ib_xrcd *xrcd;
1629 enum ib_raw_packet_caps {
1630 /* Strip cvlan from incoming packet and report it in the matching work
1631 * completion is supported.
1633 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1634 /* Scatter FCS field of an incoming packet to host memory is supported.
1636 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1637 /* Checksum offloads are supported (for both send and receive). */
1638 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1639 /* When a packet is received for an RQ with no receive WQEs, the
1640 * packet processing is delayed.
1642 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1656 struct ib_device *device;
1657 struct ib_uobject *uobject;
1659 void (*event_handler)(struct ib_event *, void *);
1663 enum ib_wq_state state;
1664 enum ib_wq_type wq_type;
1669 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1670 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1671 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1672 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1675 struct ib_wq_init_attr {
1677 enum ib_wq_type wq_type;
1681 void (*event_handler)(struct ib_event *, void *);
1682 u32 create_flags; /* Use enum ib_wq_flags */
1685 enum ib_wq_attr_mask {
1686 IB_WQ_STATE = 1 << 0,
1687 IB_WQ_CUR_STATE = 1 << 1,
1688 IB_WQ_FLAGS = 1 << 2,
1692 enum ib_wq_state wq_state;
1693 enum ib_wq_state curr_wq_state;
1694 u32 flags; /* Use enum ib_wq_flags */
1695 u32 flags_mask; /* Use enum ib_wq_flags */
1698 struct ib_rwq_ind_table {
1699 struct ib_device *device;
1700 struct ib_uobject *uobject;
1703 u32 log_ind_tbl_size;
1704 struct ib_wq **ind_tbl;
1707 struct ib_rwq_ind_table_init_attr {
1708 u32 log_ind_tbl_size;
1709 /* Each entry is a pointer to Receive Work Queue */
1710 struct ib_wq **ind_tbl;
1713 enum port_pkey_state {
1714 IB_PORT_PKEY_NOT_VALID = 0,
1715 IB_PORT_PKEY_VALID = 1,
1716 IB_PORT_PKEY_LISTED = 2,
1719 struct ib_qp_security;
1721 struct ib_port_pkey {
1722 enum port_pkey_state state;
1725 struct list_head qp_list;
1726 struct list_head to_error_list;
1727 struct ib_qp_security *sec;
1730 struct ib_ports_pkeys {
1731 struct ib_port_pkey main;
1732 struct ib_port_pkey alt;
1735 struct ib_qp_security {
1737 struct ib_device *dev;
1738 /* Hold this mutex when changing port and pkey settings. */
1740 struct ib_ports_pkeys *ports_pkeys;
1741 /* A list of all open shared QP handles. Required to enforce security
1742 * properly for all users of a shared QP.
1744 struct list_head shared_qp_list;
1747 atomic_t error_list_count;
1748 struct completion error_complete;
1749 int error_comps_pending;
1753 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1754 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1757 struct ib_device *device;
1759 struct ib_cq *send_cq;
1760 struct ib_cq *recv_cq;
1763 struct list_head rdma_mrs;
1764 struct list_head sig_mrs;
1766 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1767 struct list_head xrcd_list;
1769 /* count times opened, mcast attaches, flow attaches */
1771 struct list_head open_list;
1772 struct ib_qp *real_qp;
1773 struct ib_uobject *uobject;
1774 void (*event_handler)(struct ib_event *, void *);
1776 /* sgid_attrs associated with the AV's */
1777 const struct ib_gid_attr *av_sgid_attr;
1778 const struct ib_gid_attr *alt_path_sgid_attr;
1782 enum ib_qp_type qp_type;
1783 struct ib_rwq_ind_table *rwq_ind_tbl;
1784 struct ib_qp_security *qp_sec;
1788 * Implementation details of the RDMA core, don't use in drivers:
1790 struct rdma_restrack_entry res;
1794 struct ib_device *device;
1797 struct ib_uobject *uobject;
1802 struct ib_device *device;
1808 unsigned int page_size;
1811 struct ib_uobject *uobject; /* user */
1812 struct list_head qp_entry; /* FR */
1818 * Implementation details of the RDMA core, don't use in drivers:
1820 struct rdma_restrack_entry res;
1824 struct ib_device *device;
1826 struct ib_uobject *uobject;
1828 enum ib_mw_type type;
1832 struct ib_device *device;
1834 struct list_head list;
1839 /* Supported steering options */
1840 enum ib_flow_attr_type {
1841 /* steering according to rule specifications */
1842 IB_FLOW_ATTR_NORMAL = 0x0,
1843 /* default unicast and multicast rule -
1844 * receive all Eth traffic which isn't steered to any QP
1846 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1847 /* default multicast rule -
1848 * receive all Eth multicast traffic which isn't steered to any QP
1850 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1851 /* sniffer rule - receive all port traffic */
1852 IB_FLOW_ATTR_SNIFFER = 0x3
1855 /* Supported steering header types */
1856 enum ib_flow_spec_type {
1858 IB_FLOW_SPEC_ETH = 0x20,
1859 IB_FLOW_SPEC_IB = 0x22,
1861 IB_FLOW_SPEC_IPV4 = 0x30,
1862 IB_FLOW_SPEC_IPV6 = 0x31,
1863 IB_FLOW_SPEC_ESP = 0x34,
1865 IB_FLOW_SPEC_TCP = 0x40,
1866 IB_FLOW_SPEC_UDP = 0x41,
1867 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1868 IB_FLOW_SPEC_GRE = 0x51,
1869 IB_FLOW_SPEC_MPLS = 0x60,
1870 IB_FLOW_SPEC_INNER = 0x100,
1872 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1873 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1874 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1875 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1877 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1878 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1880 /* Flow steering rule priority is set according to it's domain.
1881 * Lower domain value means higher priority.
1883 enum ib_flow_domain {
1884 IB_FLOW_DOMAIN_USER,
1885 IB_FLOW_DOMAIN_ETHTOOL,
1888 IB_FLOW_DOMAIN_NUM /* Must be last */
1891 enum ib_flow_flags {
1892 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1893 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1894 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1897 struct ib_flow_eth_filter {
1906 struct ib_flow_spec_eth {
1909 struct ib_flow_eth_filter val;
1910 struct ib_flow_eth_filter mask;
1913 struct ib_flow_ib_filter {
1920 struct ib_flow_spec_ib {
1923 struct ib_flow_ib_filter val;
1924 struct ib_flow_ib_filter mask;
1927 /* IPv4 header flags */
1928 enum ib_ipv4_flags {
1929 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1930 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1931 last have this flag set */
1934 struct ib_flow_ipv4_filter {
1945 struct ib_flow_spec_ipv4 {
1948 struct ib_flow_ipv4_filter val;
1949 struct ib_flow_ipv4_filter mask;
1952 struct ib_flow_ipv6_filter {
1963 struct ib_flow_spec_ipv6 {
1966 struct ib_flow_ipv6_filter val;
1967 struct ib_flow_ipv6_filter mask;
1970 struct ib_flow_tcp_udp_filter {
1977 struct ib_flow_spec_tcp_udp {
1980 struct ib_flow_tcp_udp_filter val;
1981 struct ib_flow_tcp_udp_filter mask;
1984 struct ib_flow_tunnel_filter {
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990 * the tunnel_id from val has the vni value
1992 struct ib_flow_spec_tunnel {
1995 struct ib_flow_tunnel_filter val;
1996 struct ib_flow_tunnel_filter mask;
1999 struct ib_flow_esp_filter {
2006 struct ib_flow_spec_esp {
2009 struct ib_flow_esp_filter val;
2010 struct ib_flow_esp_filter mask;
2013 struct ib_flow_gre_filter {
2014 __be16 c_ks_res0_ver;
2021 struct ib_flow_spec_gre {
2024 struct ib_flow_gre_filter val;
2025 struct ib_flow_gre_filter mask;
2028 struct ib_flow_mpls_filter {
2034 struct ib_flow_spec_mpls {
2037 struct ib_flow_mpls_filter val;
2038 struct ib_flow_mpls_filter mask;
2041 struct ib_flow_spec_action_tag {
2042 enum ib_flow_spec_type type;
2047 struct ib_flow_spec_action_drop {
2048 enum ib_flow_spec_type type;
2052 struct ib_flow_spec_action_handle {
2053 enum ib_flow_spec_type type;
2055 struct ib_flow_action *act;
2058 enum ib_counters_description {
2063 struct ib_flow_spec_action_count {
2064 enum ib_flow_spec_type type;
2066 struct ib_counters *counters;
2069 union ib_flow_spec {
2074 struct ib_flow_spec_eth eth;
2075 struct ib_flow_spec_ib ib;
2076 struct ib_flow_spec_ipv4 ipv4;
2077 struct ib_flow_spec_tcp_udp tcp_udp;
2078 struct ib_flow_spec_ipv6 ipv6;
2079 struct ib_flow_spec_tunnel tunnel;
2080 struct ib_flow_spec_esp esp;
2081 struct ib_flow_spec_gre gre;
2082 struct ib_flow_spec_mpls mpls;
2083 struct ib_flow_spec_action_tag flow_tag;
2084 struct ib_flow_spec_action_drop drop;
2085 struct ib_flow_spec_action_handle action;
2086 struct ib_flow_spec_action_count flow_count;
2089 struct ib_flow_attr {
2090 enum ib_flow_attr_type type;
2096 union ib_flow_spec flows[];
2101 struct ib_device *device;
2102 struct ib_uobject *uobject;
2105 enum ib_flow_action_type {
2106 IB_FLOW_ACTION_UNSPECIFIED,
2107 IB_FLOW_ACTION_ESP = 1,
2110 struct ib_flow_action_attrs_esp_keymats {
2111 enum ib_uverbs_flow_action_esp_keymat protocol;
2113 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2117 struct ib_flow_action_attrs_esp_replays {
2118 enum ib_uverbs_flow_action_esp_replay protocol;
2120 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2124 enum ib_flow_action_attrs_esp_flags {
2125 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2126 * This is done in order to share the same flags between user-space and
2127 * kernel and spare an unnecessary translation.
2131 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2132 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2135 struct ib_flow_spec_list {
2136 struct ib_flow_spec_list *next;
2137 union ib_flow_spec spec;
2140 struct ib_flow_action_attrs_esp {
2141 struct ib_flow_action_attrs_esp_keymats *keymat;
2142 struct ib_flow_action_attrs_esp_replays *replay;
2143 struct ib_flow_spec_list *encap;
2144 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2145 * Value of 0 is a valid value.
2151 /* Use enum ib_flow_action_attrs_esp_flags */
2153 u64 hard_limit_pkts;
2156 struct ib_flow_action {
2157 struct ib_device *device;
2158 struct ib_uobject *uobject;
2159 enum ib_flow_action_type type;
2166 enum ib_process_mad_flags {
2167 IB_MAD_IGNORE_MKEY = 1,
2168 IB_MAD_IGNORE_BKEY = 2,
2169 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2172 enum ib_mad_result {
2173 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2174 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2175 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2176 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2179 struct ib_port_cache {
2181 struct ib_pkey_cache *pkey;
2182 struct ib_gid_table *gid;
2184 enum ib_port_state port_state;
2189 struct ib_event_handler event_handler;
2190 struct ib_port_cache *ports;
2195 struct ib_port_immutable {
2202 /* rdma netdev type - specifies protocol type */
2203 enum rdma_netdev_t {
2204 RDMA_NETDEV_OPA_VNIC,
2209 * struct rdma_netdev - rdma netdev
2210 * For cases where netstack interfacing is required.
2212 struct rdma_netdev {
2214 struct ib_device *hca;
2218 * cleanup function must be specified.
2219 * FIXME: This is only used for OPA_VNIC and that usage should be
2222 void (*free_rdma_netdev)(struct net_device *netdev);
2224 /* control functions */
2225 void (*set_id)(struct net_device *netdev, int id);
2227 int (*send)(struct net_device *dev, struct sk_buff *skb,
2228 struct ib_ah *address, u32 dqpn);
2230 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2231 union ib_gid *gid, u16 mlid,
2232 int set_qkey, u32 qkey);
2233 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2234 union ib_gid *gid, u16 mlid);
2237 struct ib_port_pkey_list {
2238 /* Lock to hold while modifying the list. */
2239 spinlock_t list_lock;
2240 struct list_head pkey_list;
2243 struct ib_counters {
2244 struct ib_device *device;
2245 struct ib_uobject *uobject;
2246 /* num of objects attached */
2250 struct ib_counters_read_attr {
2253 u32 flags; /* use enum ib_read_counters_flags */
2256 struct uverbs_attr_bundle;
2259 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2260 struct device *dma_device;
2262 char name[IB_DEVICE_NAME_MAX];
2264 struct list_head event_handler_list;
2265 spinlock_t event_handler_lock;
2267 rwlock_t client_data_lock;
2268 struct list_head core_list;
2269 /* Access to the client_data_list is protected by the client_data_lock
2270 * rwlock and the lists_rwsem read-write semaphore
2272 struct list_head client_data_list;
2274 struct ib_cache cache;
2276 * port_immutable is indexed by port number
2278 struct ib_port_immutable *port_immutable;
2280 int num_comp_vectors;
2282 struct ib_port_pkey_list *port_pkey_list;
2284 struct iw_cm_verbs *iwcm;
2287 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2288 * driver initialized data. The struct is kfree()'ed by the sysfs
2289 * core when the device is removed. A lifespan of -1 in the return
2290 * struct tells the core to set a default lifespan.
2292 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2295 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2296 * @index - The index in the value array we wish to have updated, or
2297 * num_counters if we want all stats updated
2299 * < 0 - Error, no counters updated
2300 * index - Updated the single counter pointed to by index
2301 * num_counters - Updated all counters (will reset the timestamp
2302 * and prevent further calls for lifespan milliseconds)
2303 * Drivers are allowed to update all counters in leiu of just the
2304 * one given in index at their option
2306 int (*get_hw_stats)(struct ib_device *device,
2307 struct rdma_hw_stats *stats,
2308 u8 port, int index);
2309 int (*query_device)(struct ib_device *device,
2310 struct ib_device_attr *device_attr,
2311 struct ib_udata *udata);
2312 int (*query_port)(struct ib_device *device,
2314 struct ib_port_attr *port_attr);
2315 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2317 /* When calling get_netdev, the HW vendor's driver should return the
2318 * net device of device @device at port @port_num or NULL if such
2319 * a net device doesn't exist. The vendor driver should call dev_hold
2320 * on this net device. The HW vendor's device driver must guarantee
2321 * that this function returns NULL before the net device has finished
2322 * NETDEV_UNREGISTER state.
2324 struct net_device *(*get_netdev)(struct ib_device *device,
2326 /* query_gid should be return GID value for @device, when @port_num
2327 * link layer is either IB or iWarp. It is no-op if @port_num port
2328 * is RoCE link layer.
2330 int (*query_gid)(struct ib_device *device,
2331 u8 port_num, int index,
2333 /* When calling add_gid, the HW vendor's driver should add the gid
2334 * of device of port at gid index available at @attr. Meta-info of
2335 * that gid (for example, the network device related to this gid) is
2336 * available at @attr. @context allows the HW vendor driver to store
2337 * extra information together with a GID entry. The HW vendor driver may
2338 * allocate memory to contain this information and store it in @context
2339 * when a new GID entry is written to. Params are consistent until the
2340 * next call of add_gid or delete_gid. The function should return 0 on
2341 * success or error otherwise. The function could be called
2342 * concurrently for different ports. This function is only called when
2343 * roce_gid_table is used.
2345 int (*add_gid)(const struct ib_gid_attr *attr,
2347 /* When calling del_gid, the HW vendor's driver should delete the
2348 * gid of device @device at gid index gid_index of port port_num
2349 * available in @attr.
2350 * Upon the deletion of a GID entry, the HW vendor must free any
2351 * allocated memory. The caller will clear @context afterwards.
2352 * This function is only called when roce_gid_table is used.
2354 int (*del_gid)(const struct ib_gid_attr *attr,
2356 int (*query_pkey)(struct ib_device *device,
2357 u8 port_num, u16 index, u16 *pkey);
2358 int (*modify_device)(struct ib_device *device,
2359 int device_modify_mask,
2360 struct ib_device_modify *device_modify);
2361 int (*modify_port)(struct ib_device *device,
2362 u8 port_num, int port_modify_mask,
2363 struct ib_port_modify *port_modify);
2364 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
2365 struct ib_udata *udata);
2366 int (*dealloc_ucontext)(struct ib_ucontext *context);
2367 int (*mmap)(struct ib_ucontext *context,
2368 struct vm_area_struct *vma);
2369 struct ib_pd * (*alloc_pd)(struct ib_device *device,
2370 struct ib_ucontext *context,
2371 struct ib_udata *udata);
2372 int (*dealloc_pd)(struct ib_pd *pd);
2373 struct ib_ah * (*create_ah)(struct ib_pd *pd,
2374 struct rdma_ah_attr *ah_attr,
2375 struct ib_udata *udata);
2376 int (*modify_ah)(struct ib_ah *ah,
2377 struct rdma_ah_attr *ah_attr);
2378 int (*query_ah)(struct ib_ah *ah,
2379 struct rdma_ah_attr *ah_attr);
2380 int (*destroy_ah)(struct ib_ah *ah);
2381 struct ib_srq * (*create_srq)(struct ib_pd *pd,
2382 struct ib_srq_init_attr *srq_init_attr,
2383 struct ib_udata *udata);
2384 int (*modify_srq)(struct ib_srq *srq,
2385 struct ib_srq_attr *srq_attr,
2386 enum ib_srq_attr_mask srq_attr_mask,
2387 struct ib_udata *udata);
2388 int (*query_srq)(struct ib_srq *srq,
2389 struct ib_srq_attr *srq_attr);
2390 int (*destroy_srq)(struct ib_srq *srq);
2391 int (*post_srq_recv)(struct ib_srq *srq,
2392 const struct ib_recv_wr *recv_wr,
2393 const struct ib_recv_wr **bad_recv_wr);
2394 struct ib_qp * (*create_qp)(struct ib_pd *pd,
2395 struct ib_qp_init_attr *qp_init_attr,
2396 struct ib_udata *udata);
2397 int (*modify_qp)(struct ib_qp *qp,
2398 struct ib_qp_attr *qp_attr,
2400 struct ib_udata *udata);
2401 int (*query_qp)(struct ib_qp *qp,
2402 struct ib_qp_attr *qp_attr,
2404 struct ib_qp_init_attr *qp_init_attr);
2405 int (*destroy_qp)(struct ib_qp *qp);
2406 int (*post_send)(struct ib_qp *qp,
2407 const struct ib_send_wr *send_wr,
2408 const struct ib_send_wr **bad_send_wr);
2409 int (*post_recv)(struct ib_qp *qp,
2410 const struct ib_recv_wr *recv_wr,
2411 const struct ib_recv_wr **bad_recv_wr);
2412 struct ib_cq * (*create_cq)(struct ib_device *device,
2413 const struct ib_cq_init_attr *attr,
2414 struct ib_ucontext *context,
2415 struct ib_udata *udata);
2416 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2418 int (*destroy_cq)(struct ib_cq *cq);
2419 int (*resize_cq)(struct ib_cq *cq, int cqe,
2420 struct ib_udata *udata);
2421 int (*poll_cq)(struct ib_cq *cq, int num_entries,
2423 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2424 int (*req_notify_cq)(struct ib_cq *cq,
2425 enum ib_cq_notify_flags flags);
2426 int (*req_ncomp_notif)(struct ib_cq *cq,
2428 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2429 int mr_access_flags);
2430 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2431 u64 start, u64 length,
2433 int mr_access_flags,
2434 struct ib_udata *udata);
2435 int (*rereg_user_mr)(struct ib_mr *mr,
2437 u64 start, u64 length,
2439 int mr_access_flags,
2441 struct ib_udata *udata);
2442 int (*dereg_mr)(struct ib_mr *mr);
2443 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2444 enum ib_mr_type mr_type,
2446 int (*map_mr_sg)(struct ib_mr *mr,
2447 struct scatterlist *sg,
2449 unsigned int *sg_offset);
2450 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2451 enum ib_mw_type type,
2452 struct ib_udata *udata);
2453 int (*dealloc_mw)(struct ib_mw *mw);
2454 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2455 int mr_access_flags,
2456 struct ib_fmr_attr *fmr_attr);
2457 int (*map_phys_fmr)(struct ib_fmr *fmr,
2458 u64 *page_list, int list_len,
2460 int (*unmap_fmr)(struct list_head *fmr_list);
2461 int (*dealloc_fmr)(struct ib_fmr *fmr);
2462 int (*attach_mcast)(struct ib_qp *qp,
2465 int (*detach_mcast)(struct ib_qp *qp,
2468 int (*process_mad)(struct ib_device *device,
2469 int process_mad_flags,
2471 const struct ib_wc *in_wc,
2472 const struct ib_grh *in_grh,
2473 const struct ib_mad_hdr *in_mad,
2475 struct ib_mad_hdr *out_mad,
2476 size_t *out_mad_size,
2477 u16 *out_mad_pkey_index);
2478 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2479 struct ib_ucontext *ucontext,
2480 struct ib_udata *udata);
2481 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2482 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2486 struct ib_udata *udata);
2487 int (*destroy_flow)(struct ib_flow *flow_id);
2488 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2489 struct ib_mr_status *mr_status);
2490 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2491 void (*drain_rq)(struct ib_qp *qp);
2492 void (*drain_sq)(struct ib_qp *qp);
2493 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2495 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2496 struct ifla_vf_info *ivf);
2497 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2498 struct ifla_vf_stats *stats);
2499 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2501 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2502 struct ib_wq_init_attr *init_attr,
2503 struct ib_udata *udata);
2504 int (*destroy_wq)(struct ib_wq *wq);
2505 int (*modify_wq)(struct ib_wq *wq,
2506 struct ib_wq_attr *attr,
2508 struct ib_udata *udata);
2509 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2510 struct ib_rwq_ind_table_init_attr *init_attr,
2511 struct ib_udata *udata);
2512 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2513 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device,
2514 const struct ib_flow_action_attrs_esp *attr,
2515 struct uverbs_attr_bundle *attrs);
2516 int (*destroy_flow_action)(struct ib_flow_action *action);
2517 int (*modify_flow_action_esp)(struct ib_flow_action *action,
2518 const struct ib_flow_action_attrs_esp *attr,
2519 struct uverbs_attr_bundle *attrs);
2520 struct ib_dm * (*alloc_dm)(struct ib_device *device,
2521 struct ib_ucontext *context,
2522 struct ib_dm_alloc_attr *attr,
2523 struct uverbs_attr_bundle *attrs);
2524 int (*dealloc_dm)(struct ib_dm *dm);
2525 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2526 struct ib_dm_mr_attr *attr,
2527 struct uverbs_attr_bundle *attrs);
2528 struct ib_counters * (*create_counters)(struct ib_device *device,
2529 struct uverbs_attr_bundle *attrs);
2530 int (*destroy_counters)(struct ib_counters *counters);
2531 int (*read_counters)(struct ib_counters *counters,
2532 struct ib_counters_read_attr *counters_read_attr,
2533 struct uverbs_attr_bundle *attrs);
2536 * rdma netdev operation
2538 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2539 * doesn't support the specified rdma netdev type.
2541 struct net_device *(*alloc_rdma_netdev)(
2542 struct ib_device *device,
2544 enum rdma_netdev_t type,
2546 unsigned char name_assign_type,
2547 void (*setup)(struct net_device *));
2549 struct module *owner;
2551 /* First group for device attributes, NULL terminated array */
2552 const struct attribute_group *groups[2];
2554 struct kobject *ports_parent;
2555 struct list_head port_list;
2558 IB_DEV_UNINITIALIZED,
2564 u64 uverbs_cmd_mask;
2565 u64 uverbs_ex_cmd_mask;
2567 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2573 struct ib_device_attr attrs;
2574 struct attribute_group *hw_stats_ag;
2575 struct rdma_hw_stats *hw_stats;
2577 #ifdef CONFIG_CGROUP_RDMA
2578 struct rdmacg_device cg_device;
2583 * Implementation details of the RDMA core, don't use in drivers
2585 struct rdma_restrack_root res;
2588 * The following mandatory functions are used only at device
2589 * registration. Keep functions such as these at the end of this
2590 * structure to avoid cache line misses when accessing struct ib_device
2593 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2594 void (*get_dev_fw_str)(struct ib_device *, char *str);
2595 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2598 const struct uverbs_object_tree_def *const *driver_specs;
2599 enum rdma_driver_id driver_id;
2604 void (*add) (struct ib_device *);
2605 void (*remove)(struct ib_device *, void *client_data);
2607 /* Returns the net_dev belonging to this ib_client and matching the
2609 * @dev: An RDMA device that the net_dev use for communication.
2610 * @port: A physical port number on the RDMA device.
2611 * @pkey: P_Key that the net_dev uses if applicable.
2612 * @gid: A GID that the net_dev uses to communicate.
2613 * @addr: An IP address the net_dev is configured with.
2614 * @client_data: The device's client data set by ib_set_client_data().
2616 * An ib_client that implements a net_dev on top of RDMA devices
2617 * (such as IP over IB) should implement this callback, allowing the
2618 * rdma_cm module to find the right net_dev for a given request.
2620 * The caller is responsible for calling dev_put on the returned
2622 struct net_device *(*get_net_dev_by_params)(
2623 struct ib_device *dev,
2626 const union ib_gid *gid,
2627 const struct sockaddr *addr,
2629 struct list_head list;
2632 struct ib_device *ib_alloc_device(size_t size);
2633 void ib_dealloc_device(struct ib_device *device);
2635 void ib_get_device_fw_str(struct ib_device *device, char *str);
2637 int ib_register_device(struct ib_device *device,
2638 int (*port_callback)(struct ib_device *,
2639 u8, struct kobject *));
2640 void ib_unregister_device(struct ib_device *device);
2642 int ib_register_client (struct ib_client *client);
2643 void ib_unregister_client(struct ib_client *client);
2645 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2646 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2649 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2650 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2651 unsigned long pfn, unsigned long size, pgprot_t prot);
2652 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2653 struct vm_area_struct *vma, struct page *page,
2654 unsigned long size);
2656 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2657 struct vm_area_struct *vma,
2658 unsigned long pfn, unsigned long size,
2663 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2664 struct vm_area_struct *vma, struct page *page,
2671 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2673 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2676 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2678 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2681 static inline bool ib_is_buffer_cleared(const void __user *p,
2687 if (len > USHRT_MAX)
2690 buf = memdup_user(p, len);
2694 ret = !memchr_inv(buf, 0, len);
2699 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2703 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2707 * ib_is_destroy_retryable - Check whether the uobject destruction
2709 * @ret: The initial destruction return code
2710 * @why: remove reason
2711 * @uobj: The uobject that is destroyed
2713 * This function is a helper function that IB layer and low-level drivers
2714 * can use to consider whether the destruction of the given uobject is
2716 * It checks the original return code, if it wasn't success the destruction
2717 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2718 * the remove reason. (i.e. why).
2719 * Must be called with the object locked for destroy.
2721 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2722 struct ib_uobject *uobj)
2724 return ret && (why == RDMA_REMOVE_DESTROY ||
2725 uobj->context->cleanup_retryable);
2729 * ib_destroy_usecnt - Called during destruction to check the usecnt
2730 * @usecnt: The usecnt atomic
2731 * @why: remove reason
2732 * @uobj: The uobject that is destroyed
2734 * Non-zero usecnts will block destruction unless destruction was triggered by
2735 * a ucontext cleanup.
2737 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2738 enum rdma_remove_reason why,
2739 struct ib_uobject *uobj)
2741 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2747 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2748 * contains all required attributes and no attributes not allowed for
2749 * the given QP state transition.
2750 * @cur_state: Current QP state
2751 * @next_state: Next QP state
2753 * @mask: Mask of supplied QP attributes
2754 * @ll : link layer of port
2756 * This function is a helper function that a low-level driver's
2757 * modify_qp method can use to validate the consumer's input. It
2758 * checks that cur_state and next_state are valid QP states, that a
2759 * transition from cur_state to next_state is allowed by the IB spec,
2760 * and that the attribute mask supplied is allowed for the transition.
2762 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2763 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2764 enum rdma_link_layer ll);
2766 void ib_register_event_handler(struct ib_event_handler *event_handler);
2767 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2768 void ib_dispatch_event(struct ib_event *event);
2770 int ib_query_port(struct ib_device *device,
2771 u8 port_num, struct ib_port_attr *port_attr);
2773 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2777 * rdma_cap_ib_switch - Check if the device is IB switch
2778 * @device: Device to check
2780 * Device driver is responsible for setting is_switch bit on
2781 * in ib_device structure at init time.
2783 * Return: true if the device is IB switch.
2785 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2787 return device->is_switch;
2791 * rdma_start_port - Return the first valid port number for the device
2794 * @device: Device to be checked
2796 * Return start port number
2798 static inline u8 rdma_start_port(const struct ib_device *device)
2800 return rdma_cap_ib_switch(device) ? 0 : 1;
2804 * rdma_end_port - Return the last valid port number for the device
2807 * @device: Device to be checked
2809 * Return last port number
2811 static inline u8 rdma_end_port(const struct ib_device *device)
2813 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2816 static inline int rdma_is_port_valid(const struct ib_device *device,
2819 return (port >= rdma_start_port(device) &&
2820 port <= rdma_end_port(device));
2823 static inline bool rdma_is_grh_required(const struct ib_device *device,
2826 return device->port_immutable[port_num].core_cap_flags &
2827 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2830 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2832 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2835 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2837 return device->port_immutable[port_num].core_cap_flags &
2838 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2841 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2843 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2846 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2848 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2851 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2853 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2856 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2858 return rdma_protocol_ib(device, port_num) ||
2859 rdma_protocol_roce(device, port_num);
2862 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2864 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2867 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2869 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2873 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2874 * Management Datagrams.
2875 * @device: Device to check
2876 * @port_num: Port number to check
2878 * Management Datagrams (MAD) are a required part of the InfiniBand
2879 * specification and are supported on all InfiniBand devices. A slightly
2880 * extended version are also supported on OPA interfaces.
2882 * Return: true if the port supports sending/receiving of MAD packets.
2884 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2886 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2890 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2891 * Management Datagrams.
2892 * @device: Device to check
2893 * @port_num: Port number to check
2895 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2896 * datagrams with their own versions. These OPA MADs share many but not all of
2897 * the characteristics of InfiniBand MADs.
2899 * OPA MADs differ in the following ways:
2901 * 1) MADs are variable size up to 2K
2902 * IBTA defined MADs remain fixed at 256 bytes
2903 * 2) OPA SMPs must carry valid PKeys
2904 * 3) OPA SMP packets are a different format
2906 * Return: true if the port supports OPA MAD packet formats.
2908 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2910 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2911 == RDMA_CORE_CAP_OPA_MAD;
2915 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2916 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2917 * @device: Device to check
2918 * @port_num: Port number to check
2920 * Each InfiniBand node is required to provide a Subnet Management Agent
2921 * that the subnet manager can access. Prior to the fabric being fully
2922 * configured by the subnet manager, the SMA is accessed via a well known
2923 * interface called the Subnet Management Interface (SMI). This interface
2924 * uses directed route packets to communicate with the SM to get around the
2925 * chicken and egg problem of the SM needing to know what's on the fabric
2926 * in order to configure the fabric, and needing to configure the fabric in
2927 * order to send packets to the devices on the fabric. These directed
2928 * route packets do not need the fabric fully configured in order to reach
2929 * their destination. The SMI is the only method allowed to send
2930 * directed route packets on an InfiniBand fabric.
2932 * Return: true if the port provides an SMI.
2934 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2936 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2940 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2941 * Communication Manager.
2942 * @device: Device to check
2943 * @port_num: Port number to check
2945 * The InfiniBand Communication Manager is one of many pre-defined General
2946 * Service Agents (GSA) that are accessed via the General Service
2947 * Interface (GSI). It's role is to facilitate establishment of connections
2948 * between nodes as well as other management related tasks for established
2951 * Return: true if the port supports an IB CM (this does not guarantee that
2952 * a CM is actually running however).
2954 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2956 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2960 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2961 * Communication Manager.
2962 * @device: Device to check
2963 * @port_num: Port number to check
2965 * Similar to above, but specific to iWARP connections which have a different
2966 * managment protocol than InfiniBand.
2968 * Return: true if the port supports an iWARP CM (this does not guarantee that
2969 * a CM is actually running however).
2971 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2973 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2977 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2978 * Subnet Administration.
2979 * @device: Device to check
2980 * @port_num: Port number to check
2982 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2983 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2984 * fabrics, devices should resolve routes to other hosts by contacting the
2985 * SA to query the proper route.
2987 * Return: true if the port should act as a client to the fabric Subnet
2988 * Administration interface. This does not imply that the SA service is
2991 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2993 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2997 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2999 * @device: Device to check
3000 * @port_num: Port number to check
3002 * InfiniBand multicast registration is more complex than normal IPv4 or
3003 * IPv6 multicast registration. Each Host Channel Adapter must register
3004 * with the Subnet Manager when it wishes to join a multicast group. It
3005 * should do so only once regardless of how many queue pairs it subscribes
3006 * to this group. And it should leave the group only after all queue pairs
3007 * attached to the group have been detached.
3009 * Return: true if the port must undertake the additional adminstrative
3010 * overhead of registering/unregistering with the SM and tracking of the
3011 * total number of queue pairs attached to the multicast group.
3013 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3015 return rdma_cap_ib_sa(device, port_num);
3019 * rdma_cap_af_ib - Check if the port of device has the capability
3020 * Native Infiniband Address.
3021 * @device: Device to check
3022 * @port_num: Port number to check
3024 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3025 * GID. RoCE uses a different mechanism, but still generates a GID via
3026 * a prescribed mechanism and port specific data.
3028 * Return: true if the port uses a GID address to identify devices on the
3031 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3033 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3037 * rdma_cap_eth_ah - Check if the port of device has the capability
3038 * Ethernet Address Handle.
3039 * @device: Device to check
3040 * @port_num: Port number to check
3042 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3043 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3044 * port. Normally, packet headers are generated by the sending host
3045 * adapter, but when sending connectionless datagrams, we must manually
3046 * inject the proper headers for the fabric we are communicating over.
3048 * Return: true if we are running as a RoCE port and must force the
3049 * addition of a Global Route Header built from our Ethernet Address
3050 * Handle into our header list for connectionless packets.
3052 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3054 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3058 * rdma_cap_opa_ah - Check if the port of device supports
3059 * OPA Address handles
3060 * @device: Device to check
3061 * @port_num: Port number to check
3063 * Return: true if we are running on an OPA device which supports
3064 * the extended OPA addressing.
3066 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3068 return (device->port_immutable[port_num].core_cap_flags &
3069 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3073 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3076 * @port_num: Port number
3078 * This MAD size includes the MAD headers and MAD payload. No other headers
3081 * Return the max MAD size required by the Port. Will return 0 if the port
3082 * does not support MADs
3084 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3086 return device->port_immutable[port_num].max_mad_size;
3090 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3091 * @device: Device to check
3092 * @port_num: Port number to check
3094 * RoCE GID table mechanism manages the various GIDs for a device.
3096 * NOTE: if allocating the port's GID table has failed, this call will still
3097 * return true, but any RoCE GID table API will fail.
3099 * Return: true if the port uses RoCE GID table mechanism in order to manage
3102 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3105 return rdma_protocol_roce(device, port_num) &&
3106 device->add_gid && device->del_gid;
3110 * Check if the device supports READ W/ INVALIDATE.
3112 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3115 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3116 * has support for it yet.
3118 return rdma_protocol_iwarp(dev, port_num);
3121 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3123 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3124 struct ifla_vf_info *info);
3125 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3126 struct ifla_vf_stats *stats);
3127 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3130 int ib_query_pkey(struct ib_device *device,
3131 u8 port_num, u16 index, u16 *pkey);
3133 int ib_modify_device(struct ib_device *device,
3134 int device_modify_mask,
3135 struct ib_device_modify *device_modify);
3137 int ib_modify_port(struct ib_device *device,
3138 u8 port_num, int port_modify_mask,
3139 struct ib_port_modify *port_modify);
3141 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3142 u8 *port_num, u16 *index);
3144 int ib_find_pkey(struct ib_device *device,
3145 u8 port_num, u16 pkey, u16 *index);
3149 * Create a memory registration for all memory in the system and place
3150 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3151 * ULPs to avoid the overhead of dynamic MRs.
3153 * This flag is generally considered unsafe and must only be used in
3154 * extremly trusted environments. Every use of it will log a warning
3155 * in the kernel log.
3157 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3160 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3161 const char *caller);
3162 #define ib_alloc_pd(device, flags) \
3163 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3164 void ib_dealloc_pd(struct ib_pd *pd);
3167 * rdma_create_ah - Creates an address handle for the given address vector.
3168 * @pd: The protection domain associated with the address handle.
3169 * @ah_attr: The attributes of the address vector.
3171 * The address handle is used to reference a local or global destination
3172 * in all UD QP post sends.
3174 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
3177 * rdma_create_user_ah - Creates an address handle for the given address vector.
3178 * It resolves destination mac address for ah attribute of RoCE type.
3179 * @pd: The protection domain associated with the address handle.
3180 * @ah_attr: The attributes of the address vector.
3181 * @udata: pointer to user's input output buffer information need by
3184 * It returns 0 on success and returns appropriate error code on error.
3185 * The address handle is used to reference a local or global destination
3186 * in all UD QP post sends.
3188 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3189 struct rdma_ah_attr *ah_attr,
3190 struct ib_udata *udata);
3192 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3194 * @hdr: the L3 header to parse
3195 * @net_type: type of header to parse
3196 * @sgid: place to store source gid
3197 * @dgid: place to store destination gid
3199 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3200 enum rdma_network_type net_type,
3201 union ib_gid *sgid, union ib_gid *dgid);
3204 * ib_get_rdma_header_version - Get the header version
3205 * @hdr: the L3 header to parse
3207 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3210 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3212 * @device: Device on which the received message arrived.
3213 * @port_num: Port on which the received message arrived.
3214 * @wc: Work completion associated with the received message.
3215 * @grh: References the received global route header. This parameter is
3216 * ignored unless the work completion indicates that the GRH is valid.
3217 * @ah_attr: Returned attributes that can be used when creating an address
3218 * handle for replying to the message.
3219 * When ib_init_ah_attr_from_wc() returns success,
3220 * (a) for IB link layer it optionally contains a reference to SGID attribute
3221 * when GRH is present for IB link layer.
3222 * (b) for RoCE link layer it contains a reference to SGID attribute.
3223 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3224 * attributes which are initialized using ib_init_ah_attr_from_wc().
3227 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3228 const struct ib_wc *wc, const struct ib_grh *grh,
3229 struct rdma_ah_attr *ah_attr);
3232 * ib_create_ah_from_wc - Creates an address handle associated with the
3233 * sender of the specified work completion.
3234 * @pd: The protection domain associated with the address handle.
3235 * @wc: Work completion information associated with a received message.
3236 * @grh: References the received global route header. This parameter is
3237 * ignored unless the work completion indicates that the GRH is valid.
3238 * @port_num: The outbound port number to associate with the address.
3240 * The address handle is used to reference a local or global destination
3241 * in all UD QP post sends.
3243 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3244 const struct ib_grh *grh, u8 port_num);
3247 * rdma_modify_ah - Modifies the address vector associated with an address
3249 * @ah: The address handle to modify.
3250 * @ah_attr: The new address vector attributes to associate with the
3253 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3256 * rdma_query_ah - Queries the address vector associated with an address
3258 * @ah: The address handle to query.
3259 * @ah_attr: The address vector attributes associated with the address
3262 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3265 * rdma_destroy_ah - Destroys an address handle.
3266 * @ah: The address handle to destroy.
3268 int rdma_destroy_ah(struct ib_ah *ah);
3271 * ib_create_srq - Creates a SRQ associated with the specified protection
3273 * @pd: The protection domain associated with the SRQ.
3274 * @srq_init_attr: A list of initial attributes required to create the
3275 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3276 * the actual capabilities of the created SRQ.
3278 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3279 * requested size of the SRQ, and set to the actual values allocated
3280 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3281 * will always be at least as large as the requested values.
3283 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3284 struct ib_srq_init_attr *srq_init_attr);
3287 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3288 * @srq: The SRQ to modify.
3289 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3290 * the current values of selected SRQ attributes are returned.
3291 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3292 * are being modified.
3294 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3295 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3296 * the number of receives queued drops below the limit.
3298 int ib_modify_srq(struct ib_srq *srq,
3299 struct ib_srq_attr *srq_attr,
3300 enum ib_srq_attr_mask srq_attr_mask);
3303 * ib_query_srq - Returns the attribute list and current values for the
3305 * @srq: The SRQ to query.
3306 * @srq_attr: The attributes of the specified SRQ.
3308 int ib_query_srq(struct ib_srq *srq,
3309 struct ib_srq_attr *srq_attr);
3312 * ib_destroy_srq - Destroys the specified SRQ.
3313 * @srq: The SRQ to destroy.
3315 int ib_destroy_srq(struct ib_srq *srq);
3318 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3319 * @srq: The SRQ to post the work request on.
3320 * @recv_wr: A list of work requests to post on the receive queue.
3321 * @bad_recv_wr: On an immediate failure, this parameter will reference
3322 * the work request that failed to be posted on the QP.
3324 static inline int ib_post_srq_recv(struct ib_srq *srq,
3325 const struct ib_recv_wr *recv_wr,
3326 const struct ib_recv_wr **bad_recv_wr)
3328 const struct ib_recv_wr *dummy;
3330 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr ? : &dummy);
3334 * ib_create_qp - Creates a QP associated with the specified protection
3336 * @pd: The protection domain associated with the QP.
3337 * @qp_init_attr: A list of initial attributes required to create the
3338 * QP. If QP creation succeeds, then the attributes are updated to
3339 * the actual capabilities of the created QP.
3341 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3342 struct ib_qp_init_attr *qp_init_attr);
3345 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3346 * @qp: The QP to modify.
3347 * @attr: On input, specifies the QP attributes to modify. On output,
3348 * the current values of selected QP attributes are returned.
3349 * @attr_mask: A bit-mask used to specify which attributes of the QP
3350 * are being modified.
3351 * @udata: pointer to user's input output buffer information
3352 * are being modified.
3353 * It returns 0 on success and returns appropriate error code on error.
3355 int ib_modify_qp_with_udata(struct ib_qp *qp,
3356 struct ib_qp_attr *attr,
3358 struct ib_udata *udata);
3361 * ib_modify_qp - Modifies the attributes for the specified QP and then
3362 * transitions the QP to the given state.
3363 * @qp: The QP to modify.
3364 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3365 * the current values of selected QP attributes are returned.
3366 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3367 * are being modified.
3369 int ib_modify_qp(struct ib_qp *qp,
3370 struct ib_qp_attr *qp_attr,
3374 * ib_query_qp - Returns the attribute list and current values for the
3376 * @qp: The QP to query.
3377 * @qp_attr: The attributes of the specified QP.
3378 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3379 * @qp_init_attr: Additional attributes of the selected QP.
3381 * The qp_attr_mask may be used to limit the query to gathering only the
3382 * selected attributes.
3384 int ib_query_qp(struct ib_qp *qp,
3385 struct ib_qp_attr *qp_attr,
3387 struct ib_qp_init_attr *qp_init_attr);
3390 * ib_destroy_qp - Destroys the specified QP.
3391 * @qp: The QP to destroy.
3393 int ib_destroy_qp(struct ib_qp *qp);
3396 * ib_open_qp - Obtain a reference to an existing sharable QP.
3397 * @xrcd - XRC domain
3398 * @qp_open_attr: Attributes identifying the QP to open.
3400 * Returns a reference to a sharable QP.
3402 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3403 struct ib_qp_open_attr *qp_open_attr);
3406 * ib_close_qp - Release an external reference to a QP.
3407 * @qp: The QP handle to release
3409 * The opened QP handle is released by the caller. The underlying
3410 * shared QP is not destroyed until all internal references are released.
3412 int ib_close_qp(struct ib_qp *qp);
3415 * ib_post_send - Posts a list of work requests to the send queue of
3417 * @qp: The QP to post the work request on.
3418 * @send_wr: A list of work requests to post on the send queue.
3419 * @bad_send_wr: On an immediate failure, this parameter will reference
3420 * the work request that failed to be posted on the QP.
3422 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3423 * error is returned, the QP state shall not be affected,
3424 * ib_post_send() will return an immediate error after queueing any
3425 * earlier work requests in the list.
3427 static inline int ib_post_send(struct ib_qp *qp,
3428 const struct ib_send_wr *send_wr,
3429 const struct ib_send_wr **bad_send_wr)
3431 const struct ib_send_wr *dummy;
3433 return qp->device->post_send(qp, send_wr, bad_send_wr ? : &dummy);
3437 * ib_post_recv - Posts a list of work requests to the receive queue of
3439 * @qp: The QP to post the work request on.
3440 * @recv_wr: A list of work requests to post on the receive queue.
3441 * @bad_recv_wr: On an immediate failure, this parameter will reference
3442 * the work request that failed to be posted on the QP.
3444 static inline int ib_post_recv(struct ib_qp *qp,
3445 const struct ib_recv_wr *recv_wr,
3446 const struct ib_recv_wr **bad_recv_wr)
3448 const struct ib_recv_wr *dummy;
3450 return qp->device->post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3453 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3454 int nr_cqe, int comp_vector,
3455 enum ib_poll_context poll_ctx, const char *caller);
3456 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3457 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3459 void ib_free_cq(struct ib_cq *cq);
3460 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3463 * ib_create_cq - Creates a CQ on the specified device.
3464 * @device: The device on which to create the CQ.
3465 * @comp_handler: A user-specified callback that is invoked when a
3466 * completion event occurs on the CQ.
3467 * @event_handler: A user-specified callback that is invoked when an
3468 * asynchronous event not associated with a completion occurs on the CQ.
3469 * @cq_context: Context associated with the CQ returned to the user via
3470 * the associated completion and event handlers.
3471 * @cq_attr: The attributes the CQ should be created upon.
3473 * Users can examine the cq structure to determine the actual CQ size.
3475 struct ib_cq *__ib_create_cq(struct ib_device *device,
3476 ib_comp_handler comp_handler,
3477 void (*event_handler)(struct ib_event *, void *),
3479 const struct ib_cq_init_attr *cq_attr,
3480 const char *caller);
3481 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3482 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3485 * ib_resize_cq - Modifies the capacity of the CQ.
3486 * @cq: The CQ to resize.
3487 * @cqe: The minimum size of the CQ.
3489 * Users can examine the cq structure to determine the actual CQ size.
3491 int ib_resize_cq(struct ib_cq *cq, int cqe);
3494 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3495 * @cq: The CQ to modify.
3496 * @cq_count: number of CQEs that will trigger an event
3497 * @cq_period: max period of time in usec before triggering an event
3500 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3503 * ib_destroy_cq - Destroys the specified CQ.
3504 * @cq: The CQ to destroy.
3506 int ib_destroy_cq(struct ib_cq *cq);
3509 * ib_poll_cq - poll a CQ for completion(s)
3510 * @cq:the CQ being polled
3511 * @num_entries:maximum number of completions to return
3512 * @wc:array of at least @num_entries &struct ib_wc where completions
3515 * Poll a CQ for (possibly multiple) completions. If the return value
3516 * is < 0, an error occurred. If the return value is >= 0, it is the
3517 * number of completions returned. If the return value is
3518 * non-negative and < num_entries, then the CQ was emptied.
3520 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3523 return cq->device->poll_cq(cq, num_entries, wc);
3527 * ib_req_notify_cq - Request completion notification on a CQ.
3528 * @cq: The CQ to generate an event for.
3530 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3531 * to request an event on the next solicited event or next work
3532 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3533 * may also be |ed in to request a hint about missed events, as
3537 * < 0 means an error occurred while requesting notification
3538 * == 0 means notification was requested successfully, and if
3539 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3540 * were missed and it is safe to wait for another event. In
3541 * this case is it guaranteed that any work completions added
3542 * to the CQ since the last CQ poll will trigger a completion
3543 * notification event.
3544 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3545 * in. It means that the consumer must poll the CQ again to
3546 * make sure it is empty to avoid missing an event because of a
3547 * race between requesting notification and an entry being
3548 * added to the CQ. This return value means it is possible
3549 * (but not guaranteed) that a work completion has been added
3550 * to the CQ since the last poll without triggering a
3551 * completion notification event.
3553 static inline int ib_req_notify_cq(struct ib_cq *cq,
3554 enum ib_cq_notify_flags flags)
3556 return cq->device->req_notify_cq(cq, flags);
3560 * ib_req_ncomp_notif - Request completion notification when there are
3561 * at least the specified number of unreaped completions on the CQ.
3562 * @cq: The CQ to generate an event for.
3563 * @wc_cnt: The number of unreaped completions that should be on the
3564 * CQ before an event is generated.
3566 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3568 return cq->device->req_ncomp_notif ?
3569 cq->device->req_ncomp_notif(cq, wc_cnt) :
3574 * ib_dma_mapping_error - check a DMA addr for error
3575 * @dev: The device for which the dma_addr was created
3576 * @dma_addr: The DMA address to check
3578 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3580 return dma_mapping_error(dev->dma_device, dma_addr);
3584 * ib_dma_map_single - Map a kernel virtual address to DMA address
3585 * @dev: The device for which the dma_addr is to be created
3586 * @cpu_addr: The kernel virtual address
3587 * @size: The size of the region in bytes
3588 * @direction: The direction of the DMA
3590 static inline u64 ib_dma_map_single(struct ib_device *dev,
3591 void *cpu_addr, size_t size,
3592 enum dma_data_direction direction)
3594 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3598 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3599 * @dev: The device for which the DMA address was created
3600 * @addr: The DMA address
3601 * @size: The size of the region in bytes
3602 * @direction: The direction of the DMA
3604 static inline void ib_dma_unmap_single(struct ib_device *dev,
3605 u64 addr, size_t size,
3606 enum dma_data_direction direction)
3608 dma_unmap_single(dev->dma_device, addr, size, direction);
3612 * ib_dma_map_page - Map a physical page to DMA address
3613 * @dev: The device for which the dma_addr is to be created
3614 * @page: The page to be mapped
3615 * @offset: The offset within the page
3616 * @size: The size of the region in bytes
3617 * @direction: The direction of the DMA
3619 static inline u64 ib_dma_map_page(struct ib_device *dev,
3621 unsigned long offset,
3623 enum dma_data_direction direction)
3625 return dma_map_page(dev->dma_device, page, offset, size, direction);
3629 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3630 * @dev: The device for which the DMA address was created
3631 * @addr: The DMA address
3632 * @size: The size of the region in bytes
3633 * @direction: The direction of the DMA
3635 static inline void ib_dma_unmap_page(struct ib_device *dev,
3636 u64 addr, size_t size,
3637 enum dma_data_direction direction)
3639 dma_unmap_page(dev->dma_device, addr, size, direction);
3643 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3644 * @dev: The device for which the DMA addresses are to be created
3645 * @sg: The array of scatter/gather entries
3646 * @nents: The number of scatter/gather entries
3647 * @direction: The direction of the DMA
3649 static inline int ib_dma_map_sg(struct ib_device *dev,
3650 struct scatterlist *sg, int nents,
3651 enum dma_data_direction direction)
3653 return dma_map_sg(dev->dma_device, sg, nents, direction);
3657 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3658 * @dev: The device for which the DMA addresses were created
3659 * @sg: The array of scatter/gather entries
3660 * @nents: The number of scatter/gather entries
3661 * @direction: The direction of the DMA
3663 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3664 struct scatterlist *sg, int nents,
3665 enum dma_data_direction direction)
3667 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3670 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3671 struct scatterlist *sg, int nents,
3672 enum dma_data_direction direction,
3673 unsigned long dma_attrs)
3675 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3679 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3680 struct scatterlist *sg, int nents,
3681 enum dma_data_direction direction,
3682 unsigned long dma_attrs)
3684 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3687 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3688 * @dev: The device for which the DMA addresses were created
3689 * @sg: The scatter/gather entry
3691 * Note: this function is obsolete. To do: change all occurrences of
3692 * ib_sg_dma_address() into sg_dma_address().
3694 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3695 struct scatterlist *sg)
3697 return sg_dma_address(sg);
3701 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3702 * @dev: The device for which the DMA addresses were created
3703 * @sg: The scatter/gather entry
3705 * Note: this function is obsolete. To do: change all occurrences of
3706 * ib_sg_dma_len() into sg_dma_len().
3708 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3709 struct scatterlist *sg)
3711 return sg_dma_len(sg);
3715 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3716 * @dev: The device for which the DMA address was created
3717 * @addr: The DMA address
3718 * @size: The size of the region in bytes
3719 * @dir: The direction of the DMA
3721 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3724 enum dma_data_direction dir)
3726 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3730 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3731 * @dev: The device for which the DMA address was created
3732 * @addr: The DMA address
3733 * @size: The size of the region in bytes
3734 * @dir: The direction of the DMA
3736 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3739 enum dma_data_direction dir)
3741 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3745 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3746 * @dev: The device for which the DMA address is requested
3747 * @size: The size of the region to allocate in bytes
3748 * @dma_handle: A pointer for returning the DMA address of the region
3749 * @flag: memory allocator flags
3751 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3753 dma_addr_t *dma_handle,
3756 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3760 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3761 * @dev: The device for which the DMA addresses were allocated
3762 * @size: The size of the region
3763 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3764 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3766 static inline void ib_dma_free_coherent(struct ib_device *dev,
3767 size_t size, void *cpu_addr,
3768 dma_addr_t dma_handle)
3770 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3774 * ib_dereg_mr - Deregisters a memory region and removes it from the
3775 * HCA translation table.
3776 * @mr: The memory region to deregister.
3778 * This function can fail, if the memory region has memory windows bound to it.
3780 int ib_dereg_mr(struct ib_mr *mr);
3782 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3783 enum ib_mr_type mr_type,
3787 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3789 * @mr - struct ib_mr pointer to be updated.
3790 * @newkey - new key to be used.
3792 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3794 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3795 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3799 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3800 * for calculating a new rkey for type 2 memory windows.
3801 * @rkey - the rkey to increment.
3803 static inline u32 ib_inc_rkey(u32 rkey)
3805 const u32 mask = 0x000000ff;
3806 return ((rkey + 1) & mask) | (rkey & ~mask);
3810 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3811 * @pd: The protection domain associated with the unmapped region.
3812 * @mr_access_flags: Specifies the memory access rights.
3813 * @fmr_attr: Attributes of the unmapped region.
3815 * A fast memory region must be mapped before it can be used as part of
3818 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3819 int mr_access_flags,
3820 struct ib_fmr_attr *fmr_attr);
3823 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3824 * @fmr: The fast memory region to associate with the pages.
3825 * @page_list: An array of physical pages to map to the fast memory region.
3826 * @list_len: The number of pages in page_list.
3827 * @iova: The I/O virtual address to use with the mapped region.
3829 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3830 u64 *page_list, int list_len,
3833 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3837 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3838 * @fmr_list: A linked list of fast memory regions to unmap.
3840 int ib_unmap_fmr(struct list_head *fmr_list);
3843 * ib_dealloc_fmr - Deallocates a fast memory region.
3844 * @fmr: The fast memory region to deallocate.
3846 int ib_dealloc_fmr(struct ib_fmr *fmr);
3849 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3850 * @qp: QP to attach to the multicast group. The QP must be type
3852 * @gid: Multicast group GID.
3853 * @lid: Multicast group LID in host byte order.
3855 * In order to send and receive multicast packets, subnet
3856 * administration must have created the multicast group and configured
3857 * the fabric appropriately. The port associated with the specified
3858 * QP must also be a member of the multicast group.
3860 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3863 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3864 * @qp: QP to detach from the multicast group.
3865 * @gid: Multicast group GID.
3866 * @lid: Multicast group LID in host byte order.
3868 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3871 * ib_alloc_xrcd - Allocates an XRC domain.
3872 * @device: The device on which to allocate the XRC domain.
3873 * @caller: Module name for kernel consumers
3875 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3876 #define ib_alloc_xrcd(device) \
3877 __ib_alloc_xrcd((device), KBUILD_MODNAME)
3880 * ib_dealloc_xrcd - Deallocates an XRC domain.
3881 * @xrcd: The XRC domain to deallocate.
3883 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3885 static inline int ib_check_mr_access(int flags)
3888 * Local write permission is required if remote write or
3889 * remote atomic permission is also requested.
3891 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3892 !(flags & IB_ACCESS_LOCAL_WRITE))
3898 static inline bool ib_access_writable(int access_flags)
3901 * We have writable memory backing the MR if any of the following
3902 * access flags are set. "Local write" and "remote write" obviously
3903 * require write access. "Remote atomic" can do things like fetch and
3904 * add, which will modify memory, and "MW bind" can change permissions
3905 * by binding a window.
3907 return access_flags &
3908 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
3909 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3913 * ib_check_mr_status: lightweight check of MR status.
3914 * This routine may provide status checks on a selected
3915 * ib_mr. first use is for signature status check.
3917 * @mr: A memory region.
3918 * @check_mask: Bitmask of which checks to perform from
3919 * ib_mr_status_check enumeration.
3920 * @mr_status: The container of relevant status checks.
3921 * failed checks will be indicated in the status bitmask
3922 * and the relevant info shall be in the error item.
3924 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3925 struct ib_mr_status *mr_status);
3927 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3928 u16 pkey, const union ib_gid *gid,
3929 const struct sockaddr *addr);
3930 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3931 struct ib_wq_init_attr *init_attr);
3932 int ib_destroy_wq(struct ib_wq *wq);
3933 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3935 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3936 struct ib_rwq_ind_table_init_attr*
3937 wq_ind_table_init_attr);
3938 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3940 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3941 unsigned int *sg_offset, unsigned int page_size);
3944 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3945 unsigned int *sg_offset, unsigned int page_size)
3949 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3955 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3956 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3958 void ib_drain_rq(struct ib_qp *qp);
3959 void ib_drain_sq(struct ib_qp *qp);
3960 void ib_drain_qp(struct ib_qp *qp);
3962 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3964 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3966 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3967 return attr->roce.dmac;
3971 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3973 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3974 attr->ib.dlid = (u16)dlid;
3975 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3976 attr->opa.dlid = dlid;
3979 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3981 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3982 return attr->ib.dlid;
3983 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3984 return attr->opa.dlid;
3988 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3993 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3998 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4001 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4002 attr->ib.src_path_bits = src_path_bits;
4003 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4004 attr->opa.src_path_bits = src_path_bits;
4007 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4009 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4010 return attr->ib.src_path_bits;
4011 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4012 return attr->opa.src_path_bits;
4016 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4019 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4020 attr->opa.make_grd = make_grd;
4023 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4025 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4026 return attr->opa.make_grd;
4030 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4032 attr->port_num = port_num;
4035 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4037 return attr->port_num;
4040 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4043 attr->static_rate = static_rate;
4046 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4048 return attr->static_rate;
4051 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4052 enum ib_ah_flags flag)
4054 attr->ah_flags = flag;
4057 static inline enum ib_ah_flags
4058 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4060 return attr->ah_flags;
4063 static inline const struct ib_global_route
4064 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4069 /*To retrieve and modify the grh */
4070 static inline struct ib_global_route
4071 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4076 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4078 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4080 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4083 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4086 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4088 grh->dgid.global.subnet_prefix = prefix;
4091 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4094 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4096 grh->dgid.global.interface_id = if_id;
4099 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4100 union ib_gid *dgid, u32 flow_label,
4101 u8 sgid_index, u8 hop_limit,
4104 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4106 attr->ah_flags = IB_AH_GRH;
4109 grh->flow_label = flow_label;
4110 grh->sgid_index = sgid_index;
4111 grh->hop_limit = hop_limit;
4112 grh->traffic_class = traffic_class;
4113 grh->sgid_attr = NULL;
4116 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4117 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4118 u32 flow_label, u8 hop_limit, u8 traffic_class,
4119 const struct ib_gid_attr *sgid_attr);
4120 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4121 const struct rdma_ah_attr *src);
4122 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4123 const struct rdma_ah_attr *new);
4124 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4127 * rdma_ah_find_type - Return address handle type.
4129 * @dev: Device to be checked
4130 * @port_num: Port number
4132 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4135 if (rdma_protocol_roce(dev, port_num))
4136 return RDMA_AH_ATTR_TYPE_ROCE;
4137 if (rdma_protocol_ib(dev, port_num)) {
4138 if (rdma_cap_opa_ah(dev, port_num))
4139 return RDMA_AH_ATTR_TYPE_OPA;
4140 return RDMA_AH_ATTR_TYPE_IB;
4143 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4147 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4148 * In the current implementation the only way to get
4149 * get the 32bit lid is from other sources for OPA.
4150 * For IB, lids will always be 16bits so cast the
4151 * value accordingly.
4155 static inline u16 ib_lid_cpu16(u32 lid)
4157 WARN_ON_ONCE(lid & 0xFFFF0000);
4162 * ib_lid_be16 - Return lid in 16bit BE encoding.
4166 static inline __be16 ib_lid_be16(u32 lid)
4168 WARN_ON_ONCE(lid & 0xFFFF0000);
4169 return cpu_to_be16((u16)lid);
4173 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4175 * @device: the rdma device
4176 * @comp_vector: index of completion vector
4178 * Returns NULL on failure, otherwise a corresponding cpu map of the
4179 * completion vector (returns all-cpus map if the device driver doesn't
4180 * implement get_vector_affinity).
4182 static inline const struct cpumask *
4183 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4185 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4186 !device->get_vector_affinity)
4189 return device->get_vector_affinity(device, comp_vector);
4194 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4195 * and add their gids, as needed, to the relevant RoCE devices.
4197 * @device: the rdma device
4199 void rdma_roce_rescan_device(struct ib_device *ibdev);
4201 struct ib_ucontext *ib_uverbs_get_ucontext(struct ib_uverbs_file *ufile);
4203 int uverbs_destroy_def_handler(struct ib_uverbs_file *file,
4204 struct uverbs_attr_bundle *attrs);
4205 #endif /* IB_VERBS_H */