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>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
72 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
76 extern struct workqueue_struct *ib_wq;
77 extern struct workqueue_struct *ib_comp_wq;
78 extern struct workqueue_struct *ib_comp_unbound_wq;
81 void ibdev_printk(const char *level, const struct ib_device *ibdev,
82 const char *format, ...);
84 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
86 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
88 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
90 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
92 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
94 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
96 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
98 #if defined(CONFIG_DYNAMIC_DEBUG)
99 #define ibdev_dbg(__dev, format, args...) \
100 dynamic_ibdev_dbg(__dev, format, ##args)
102 #define ibdev_dbg(__dev, format, args...) \
103 ibdev_printk(KERN_DEBUG, __dev, format, ##args)
105 __printf(2, 3) __cold
107 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
113 __be64 subnet_prefix;
118 extern union ib_gid zgid;
121 /* If link layer is Ethernet, this is RoCE V1 */
123 IB_GID_TYPE_ROCE = 0,
124 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
128 #define ROCE_V2_UDP_DPORT 4791
130 struct net_device __rcu *ndev;
131 struct ib_device *device;
133 enum ib_gid_type gid_type;
139 /* set the local administered indication */
140 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
143 enum rdma_transport_type {
145 RDMA_TRANSPORT_IWARP,
146 RDMA_TRANSPORT_USNIC,
147 RDMA_TRANSPORT_USNIC_UDP,
148 RDMA_TRANSPORT_UNSPECIFIED,
151 enum rdma_protocol_type {
155 RDMA_PROTOCOL_USNIC_UDP
158 __attribute_const__ enum rdma_transport_type
159 rdma_node_get_transport(unsigned int node_type);
161 enum rdma_network_type {
163 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
168 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
170 if (network_type == RDMA_NETWORK_IPV4 ||
171 network_type == RDMA_NETWORK_IPV6)
172 return IB_GID_TYPE_ROCE_UDP_ENCAP;
174 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
175 return IB_GID_TYPE_IB;
178 static inline enum rdma_network_type
179 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
181 if (attr->gid_type == IB_GID_TYPE_IB)
182 return RDMA_NETWORK_IB;
184 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
185 return RDMA_NETWORK_IPV4;
187 return RDMA_NETWORK_IPV6;
190 enum rdma_link_layer {
191 IB_LINK_LAYER_UNSPECIFIED,
192 IB_LINK_LAYER_INFINIBAND,
193 IB_LINK_LAYER_ETHERNET,
196 enum ib_device_cap_flags {
197 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
198 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
199 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
200 IB_DEVICE_RAW_MULTI = (1 << 3),
201 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
202 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
203 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
204 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
205 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
206 /* Not in use, former INIT_TYPE = (1 << 9),*/
207 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
208 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
209 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
210 IB_DEVICE_SRQ_RESIZE = (1 << 13),
211 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
214 * This device supports a per-device lkey or stag that can be
215 * used without performing a memory registration for the local
216 * memory. Note that ULPs should never check this flag, but
217 * instead of use the local_dma_lkey flag in the ib_pd structure,
218 * which will always contain a usable lkey.
220 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
221 /* Reserved, old SEND_W_INV = (1 << 16),*/
222 IB_DEVICE_MEM_WINDOW = (1 << 17),
224 * Devices should set IB_DEVICE_UD_IP_SUM if they support
225 * insertion of UDP and TCP checksum on outgoing UD IPoIB
226 * messages and can verify the validity of checksum for
227 * incoming messages. Setting this flag implies that the
228 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
230 IB_DEVICE_UD_IP_CSUM = (1 << 18),
231 IB_DEVICE_UD_TSO = (1 << 19),
232 IB_DEVICE_XRC = (1 << 20),
235 * This device supports the IB "base memory management extension",
236 * which includes support for fast registrations (IB_WR_REG_MR,
237 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
238 * also be set by any iWarp device which must support FRs to comply
239 * to the iWarp verbs spec. iWarp devices also support the
240 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
243 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
244 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
245 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
246 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
247 IB_DEVICE_RC_IP_CSUM = (1 << 25),
248 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
249 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
251 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
252 * support execution of WQEs that involve synchronization
253 * of I/O operations with single completion queue managed
256 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
257 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
258 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
259 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
260 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
261 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
262 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
263 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
264 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
265 /* The device supports padding incoming writes to cacheline. */
266 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
267 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
276 enum ib_odp_general_cap_bits {
277 IB_ODP_SUPPORT = 1 << 0,
278 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
281 enum ib_odp_transport_cap_bits {
282 IB_ODP_SUPPORT_SEND = 1 << 0,
283 IB_ODP_SUPPORT_RECV = 1 << 1,
284 IB_ODP_SUPPORT_WRITE = 1 << 2,
285 IB_ODP_SUPPORT_READ = 1 << 3,
286 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
287 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
291 uint64_t general_caps;
293 uint32_t rc_odp_caps;
294 uint32_t uc_odp_caps;
295 uint32_t ud_odp_caps;
296 uint32_t xrc_odp_caps;
297 } per_transport_caps;
301 /* Corresponding bit will be set if qp type from
302 * 'enum ib_qp_type' is supported, e.g.
303 * supported_qpts |= 1 << IB_QPT_UD
306 u32 max_rwq_indirection_tables;
307 u32 max_rwq_indirection_table_size;
310 enum ib_tm_cap_flags {
311 /* Support tag matching with rendezvous offload for RC transport */
312 IB_TM_CAP_RNDV_RC = 1 << 0,
316 /* Max size of RNDV header */
317 u32 max_rndv_hdr_size;
318 /* Max number of entries in tag matching list */
320 /* From enum ib_tm_cap_flags */
322 /* Max number of outstanding list operations */
324 /* Max number of SGE in tag matching entry */
328 struct ib_cq_init_attr {
334 enum ib_cq_attr_mask {
335 IB_CQ_MODERATE = 1 << 0,
339 u16 max_cq_moderation_count;
340 u16 max_cq_moderation_period;
343 struct ib_dm_mr_attr {
349 struct ib_dm_alloc_attr {
355 struct ib_device_attr {
357 __be64 sys_image_guid;
365 u64 device_cap_flags;
376 int max_qp_init_rd_atom;
377 int max_ee_init_rd_atom;
378 enum ib_atomic_cap atomic_cap;
379 enum ib_atomic_cap masked_atomic_cap;
386 int max_mcast_qp_attach;
387 int max_total_mcast_qp_attach;
394 unsigned int max_fast_reg_page_list_len;
395 unsigned int max_pi_fast_reg_page_list_len;
397 u8 local_ca_ack_delay;
400 struct ib_odp_caps odp_caps;
401 uint64_t timestamp_mask;
402 uint64_t hca_core_clock; /* in KHZ */
403 struct ib_rss_caps rss_caps;
405 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
406 struct ib_tm_caps tm_caps;
407 struct ib_cq_caps cq_caps;
419 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
422 case IB_MTU_256: return 256;
423 case IB_MTU_512: return 512;
424 case IB_MTU_1024: return 1024;
425 case IB_MTU_2048: return 2048;
426 case IB_MTU_4096: return 4096;
431 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
435 else if (mtu >= 2048)
437 else if (mtu >= 1024)
451 IB_PORT_ACTIVE_DEFER = 5
462 static inline int ib_width_enum_to_int(enum ib_port_width width)
465 case IB_WIDTH_1X: return 1;
466 case IB_WIDTH_2X: return 2;
467 case IB_WIDTH_4X: return 4;
468 case IB_WIDTH_8X: return 8;
469 case IB_WIDTH_12X: return 12;
485 * struct rdma_hw_stats
486 * @lock - Mutex to protect parallel write access to lifespan and values
487 * of counters, which are 64bits and not guaranteeed to be written
488 * atomicaly on 32bits systems.
489 * @timestamp - Used by the core code to track when the last update was
490 * @lifespan - Used by the core code to determine how old the counters
491 * should be before being updated again. Stored in jiffies, defaults
492 * to 10 milliseconds, drivers can override the default be specifying
493 * their own value during their allocation routine.
494 * @name - Array of pointers to static names used for the counters in
496 * @num_counters - How many hardware counters there are. If name is
497 * shorter than this number, a kernel oops will result. Driver authors
498 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
499 * in their code to prevent this.
500 * @value - Array of u64 counters that are accessed by the sysfs code and
501 * filled in by the drivers get_stats routine
503 struct rdma_hw_stats {
504 struct mutex lock; /* Protect lifespan and values[] */
505 unsigned long timestamp;
506 unsigned long lifespan;
507 const char * const *names;
512 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
514 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
516 * @names - Array of static const char *
517 * @num_counters - How many elements in array
518 * @lifespan - How many milliseconds between updates
520 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
521 const char * const *names, int num_counters,
522 unsigned long lifespan)
524 struct rdma_hw_stats *stats;
526 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
530 stats->names = names;
531 stats->num_counters = num_counters;
532 stats->lifespan = msecs_to_jiffies(lifespan);
538 /* Define bits for the various functionality this port needs to be supported by
541 /* Management 0x00000FFF */
542 #define RDMA_CORE_CAP_IB_MAD 0x00000001
543 #define RDMA_CORE_CAP_IB_SMI 0x00000002
544 #define RDMA_CORE_CAP_IB_CM 0x00000004
545 #define RDMA_CORE_CAP_IW_CM 0x00000008
546 #define RDMA_CORE_CAP_IB_SA 0x00000010
547 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
549 /* Address format 0x000FF000 */
550 #define RDMA_CORE_CAP_AF_IB 0x00001000
551 #define RDMA_CORE_CAP_ETH_AH 0x00002000
552 #define RDMA_CORE_CAP_OPA_AH 0x00004000
553 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
555 /* Protocol 0xFFF00000 */
556 #define RDMA_CORE_CAP_PROT_IB 0x00100000
557 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
558 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
559 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
560 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
561 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
563 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
564 | RDMA_CORE_CAP_PROT_ROCE \
565 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
567 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
568 | RDMA_CORE_CAP_IB_MAD \
569 | RDMA_CORE_CAP_IB_SMI \
570 | RDMA_CORE_CAP_IB_CM \
571 | RDMA_CORE_CAP_IB_SA \
572 | RDMA_CORE_CAP_AF_IB)
573 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
574 | RDMA_CORE_CAP_IB_MAD \
575 | RDMA_CORE_CAP_IB_CM \
576 | RDMA_CORE_CAP_AF_IB \
577 | RDMA_CORE_CAP_ETH_AH)
578 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
579 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
580 | RDMA_CORE_CAP_IB_MAD \
581 | RDMA_CORE_CAP_IB_CM \
582 | RDMA_CORE_CAP_AF_IB \
583 | RDMA_CORE_CAP_ETH_AH)
584 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
585 | RDMA_CORE_CAP_IW_CM)
586 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
587 | RDMA_CORE_CAP_OPA_MAD)
589 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
591 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
593 struct ib_port_attr {
595 enum ib_port_state state;
597 enum ib_mtu active_mtu;
599 unsigned int ip_gids:1;
600 /* This is the value from PortInfo CapabilityMask, defined by IBA */
619 enum ib_device_modify_flags {
620 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
621 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
624 #define IB_DEVICE_NODE_DESC_MAX 64
626 struct ib_device_modify {
628 char node_desc[IB_DEVICE_NODE_DESC_MAX];
631 enum ib_port_modify_flags {
632 IB_PORT_SHUTDOWN = 1,
633 IB_PORT_INIT_TYPE = (1<<2),
634 IB_PORT_RESET_QKEY_CNTR = (1<<3),
635 IB_PORT_OPA_MASK_CHG = (1<<4)
638 struct ib_port_modify {
639 u32 set_port_cap_mask;
640 u32 clr_port_cap_mask;
648 IB_EVENT_QP_ACCESS_ERR,
652 IB_EVENT_PATH_MIG_ERR,
653 IB_EVENT_DEVICE_FATAL,
654 IB_EVENT_PORT_ACTIVE,
657 IB_EVENT_PKEY_CHANGE,
660 IB_EVENT_SRQ_LIMIT_REACHED,
661 IB_EVENT_QP_LAST_WQE_REACHED,
662 IB_EVENT_CLIENT_REREGISTER,
667 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
670 struct ib_device *device;
678 enum ib_event_type event;
681 struct ib_event_handler {
682 struct ib_device *device;
683 void (*handler)(struct ib_event_handler *, struct ib_event *);
684 struct list_head list;
687 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
689 (_ptr)->device = _device; \
690 (_ptr)->handler = _handler; \
691 INIT_LIST_HEAD(&(_ptr)->list); \
694 struct ib_global_route {
695 const struct ib_gid_attr *sgid_attr;
704 __be32 version_tclass_flow;
712 union rdma_network_hdr {
715 /* The IB spec states that if it's IPv4, the header
716 * is located in the last 20 bytes of the header.
719 struct iphdr roce4grh;
723 #define IB_QPN_MASK 0xFFFFFF
726 IB_MULTICAST_QPN = 0xffffff
729 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
730 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
737 IB_RATE_PORT_CURRENT = 0,
738 IB_RATE_2_5_GBPS = 2,
746 IB_RATE_120_GBPS = 10,
747 IB_RATE_14_GBPS = 11,
748 IB_RATE_56_GBPS = 12,
749 IB_RATE_112_GBPS = 13,
750 IB_RATE_168_GBPS = 14,
751 IB_RATE_25_GBPS = 15,
752 IB_RATE_100_GBPS = 16,
753 IB_RATE_200_GBPS = 17,
754 IB_RATE_300_GBPS = 18,
755 IB_RATE_28_GBPS = 19,
756 IB_RATE_50_GBPS = 20,
757 IB_RATE_400_GBPS = 21,
758 IB_RATE_600_GBPS = 22,
762 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
763 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
764 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
765 * @rate: rate to convert.
767 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
770 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
771 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
772 * @rate: rate to convert.
774 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
778 * enum ib_mr_type - memory region type
779 * @IB_MR_TYPE_MEM_REG: memory region that is used for
780 * normal registration
781 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
782 * register any arbitrary sg lists (without
783 * the normal mr constraints - see
785 * @IB_MR_TYPE_DM: memory region that is used for device
786 * memory registration
787 * @IB_MR_TYPE_USER: memory region that is used for the user-space
789 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
790 * without address translations (VA=PA)
791 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
792 * data integrity operations
800 IB_MR_TYPE_INTEGRITY,
803 enum ib_mr_status_check {
804 IB_MR_CHECK_SIG_STATUS = 1,
808 * struct ib_mr_status - Memory region status container
810 * @fail_status: Bitmask of MR checks status. For each
811 * failed check a corresponding status bit is set.
812 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
815 struct ib_mr_status {
817 struct ib_sig_err sig_err;
821 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
823 * @mult: multiple to convert.
825 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
827 enum rdma_ah_attr_type {
828 RDMA_AH_ATTR_TYPE_UNDEFINED,
829 RDMA_AH_ATTR_TYPE_IB,
830 RDMA_AH_ATTR_TYPE_ROCE,
831 RDMA_AH_ATTR_TYPE_OPA,
839 struct roce_ah_attr {
849 struct rdma_ah_attr {
850 struct ib_global_route grh;
855 enum rdma_ah_attr_type type;
857 struct ib_ah_attr ib;
858 struct roce_ah_attr roce;
859 struct opa_ah_attr opa;
867 IB_WC_LOC_EEC_OP_ERR,
872 IB_WC_LOC_ACCESS_ERR,
873 IB_WC_REM_INV_REQ_ERR,
874 IB_WC_REM_ACCESS_ERR,
877 IB_WC_RNR_RETRY_EXC_ERR,
878 IB_WC_LOC_RDD_VIOL_ERR,
879 IB_WC_REM_INV_RD_REQ_ERR,
882 IB_WC_INV_EEC_STATE_ERR,
884 IB_WC_RESP_TIMEOUT_ERR,
888 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
899 IB_WC_MASKED_COMP_SWAP,
900 IB_WC_MASKED_FETCH_ADD,
902 * Set value of IB_WC_RECV so consumers can test if a completion is a
903 * receive by testing (opcode & IB_WC_RECV).
906 IB_WC_RECV_RDMA_WITH_IMM
911 IB_WC_WITH_IMM = (1<<1),
912 IB_WC_WITH_INVALIDATE = (1<<2),
913 IB_WC_IP_CSUM_OK = (1<<3),
914 IB_WC_WITH_SMAC = (1<<4),
915 IB_WC_WITH_VLAN = (1<<5),
916 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
922 struct ib_cqe *wr_cqe;
924 enum ib_wc_status status;
925 enum ib_wc_opcode opcode;
939 u8 port_num; /* valid only for DR SMPs on switches */
945 enum ib_cq_notify_flags {
946 IB_CQ_SOLICITED = 1 << 0,
947 IB_CQ_NEXT_COMP = 1 << 1,
948 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
949 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
958 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
960 return srq_type == IB_SRQT_XRC ||
961 srq_type == IB_SRQT_TM;
964 enum ib_srq_attr_mask {
965 IB_SRQ_MAX_WR = 1 << 0,
966 IB_SRQ_LIMIT = 1 << 1,
975 struct ib_srq_init_attr {
976 void (*event_handler)(struct ib_event *, void *);
978 struct ib_srq_attr attr;
979 enum ib_srq_type srq_type;
985 struct ib_xrcd *xrcd;
1000 u32 max_inline_data;
1003 * Maximum number of rdma_rw_ctx structures in flight at a time.
1004 * ib_create_qp() will calculate the right amount of neededed WRs
1005 * and MRs based on this.
1017 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1018 * here (and in that order) since the MAD layer uses them as
1019 * indices into a 2-entry table.
1028 IB_QPT_RAW_ETHERTYPE,
1029 IB_QPT_RAW_PACKET = 8,
1033 IB_QPT_DRIVER = 0xFF,
1034 /* Reserve a range for qp types internal to the low level driver.
1035 * These qp types will not be visible at the IB core layer, so the
1036 * IB_QPT_MAX usages should not be affected in the core layer
1038 IB_QPT_RESERVED1 = 0x1000,
1050 enum ib_qp_create_flags {
1051 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1052 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1053 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1054 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1055 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1056 IB_QP_CREATE_NETIF_QP = 1 << 5,
1057 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1058 /* FREE = 1 << 7, */
1059 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1060 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1061 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1062 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1063 /* reserve bits 26-31 for low level drivers' internal use */
1064 IB_QP_CREATE_RESERVED_START = 1 << 26,
1065 IB_QP_CREATE_RESERVED_END = 1 << 31,
1069 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1070 * callback to destroy the passed in QP.
1073 struct ib_qp_init_attr {
1074 /* Consumer's event_handler callback must not block */
1075 void (*event_handler)(struct ib_event *, void *);
1078 struct ib_cq *send_cq;
1079 struct ib_cq *recv_cq;
1081 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1082 struct ib_qp_cap cap;
1083 enum ib_sig_type sq_sig_type;
1084 enum ib_qp_type qp_type;
1088 * Only needed for special QP types, or when using the RW API.
1091 struct ib_rwq_ind_table *rwq_ind_tbl;
1095 struct ib_qp_open_attr {
1096 void (*event_handler)(struct ib_event *, void *);
1099 enum ib_qp_type qp_type;
1102 enum ib_rnr_timeout {
1103 IB_RNR_TIMER_655_36 = 0,
1104 IB_RNR_TIMER_000_01 = 1,
1105 IB_RNR_TIMER_000_02 = 2,
1106 IB_RNR_TIMER_000_03 = 3,
1107 IB_RNR_TIMER_000_04 = 4,
1108 IB_RNR_TIMER_000_06 = 5,
1109 IB_RNR_TIMER_000_08 = 6,
1110 IB_RNR_TIMER_000_12 = 7,
1111 IB_RNR_TIMER_000_16 = 8,
1112 IB_RNR_TIMER_000_24 = 9,
1113 IB_RNR_TIMER_000_32 = 10,
1114 IB_RNR_TIMER_000_48 = 11,
1115 IB_RNR_TIMER_000_64 = 12,
1116 IB_RNR_TIMER_000_96 = 13,
1117 IB_RNR_TIMER_001_28 = 14,
1118 IB_RNR_TIMER_001_92 = 15,
1119 IB_RNR_TIMER_002_56 = 16,
1120 IB_RNR_TIMER_003_84 = 17,
1121 IB_RNR_TIMER_005_12 = 18,
1122 IB_RNR_TIMER_007_68 = 19,
1123 IB_RNR_TIMER_010_24 = 20,
1124 IB_RNR_TIMER_015_36 = 21,
1125 IB_RNR_TIMER_020_48 = 22,
1126 IB_RNR_TIMER_030_72 = 23,
1127 IB_RNR_TIMER_040_96 = 24,
1128 IB_RNR_TIMER_061_44 = 25,
1129 IB_RNR_TIMER_081_92 = 26,
1130 IB_RNR_TIMER_122_88 = 27,
1131 IB_RNR_TIMER_163_84 = 28,
1132 IB_RNR_TIMER_245_76 = 29,
1133 IB_RNR_TIMER_327_68 = 30,
1134 IB_RNR_TIMER_491_52 = 31
1137 enum ib_qp_attr_mask {
1139 IB_QP_CUR_STATE = (1<<1),
1140 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1141 IB_QP_ACCESS_FLAGS = (1<<3),
1142 IB_QP_PKEY_INDEX = (1<<4),
1143 IB_QP_PORT = (1<<5),
1144 IB_QP_QKEY = (1<<6),
1146 IB_QP_PATH_MTU = (1<<8),
1147 IB_QP_TIMEOUT = (1<<9),
1148 IB_QP_RETRY_CNT = (1<<10),
1149 IB_QP_RNR_RETRY = (1<<11),
1150 IB_QP_RQ_PSN = (1<<12),
1151 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1152 IB_QP_ALT_PATH = (1<<14),
1153 IB_QP_MIN_RNR_TIMER = (1<<15),
1154 IB_QP_SQ_PSN = (1<<16),
1155 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1156 IB_QP_PATH_MIG_STATE = (1<<18),
1157 IB_QP_CAP = (1<<19),
1158 IB_QP_DEST_QPN = (1<<20),
1159 IB_QP_RESERVED1 = (1<<21),
1160 IB_QP_RESERVED2 = (1<<22),
1161 IB_QP_RESERVED3 = (1<<23),
1162 IB_QP_RESERVED4 = (1<<24),
1163 IB_QP_RATE_LIMIT = (1<<25),
1188 enum ib_qp_state qp_state;
1189 enum ib_qp_state cur_qp_state;
1190 enum ib_mtu path_mtu;
1191 enum ib_mig_state path_mig_state;
1196 int qp_access_flags;
1197 struct ib_qp_cap cap;
1198 struct rdma_ah_attr ah_attr;
1199 struct rdma_ah_attr alt_ah_attr;
1202 u8 en_sqd_async_notify;
1205 u8 max_dest_rd_atomic;
1217 /* These are shared with userspace */
1218 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1219 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1220 IB_WR_SEND = IB_UVERBS_WR_SEND,
1221 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1222 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1223 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1224 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1225 IB_WR_LSO = IB_UVERBS_WR_TSO,
1226 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1227 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1228 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1229 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1230 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1231 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1232 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1234 /* These are kernel only and can not be issued by userspace */
1235 IB_WR_REG_MR = 0x20,
1236 IB_WR_REG_MR_INTEGRITY,
1238 /* reserve values for low level drivers' internal use.
1239 * These values will not be used at all in the ib core layer.
1241 IB_WR_RESERVED1 = 0xf0,
1253 enum ib_send_flags {
1255 IB_SEND_SIGNALED = (1<<1),
1256 IB_SEND_SOLICITED = (1<<2),
1257 IB_SEND_INLINE = (1<<3),
1258 IB_SEND_IP_CSUM = (1<<4),
1260 /* reserve bits 26-31 for low level drivers' internal use */
1261 IB_SEND_RESERVED_START = (1 << 26),
1262 IB_SEND_RESERVED_END = (1 << 31),
1272 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1276 struct ib_send_wr *next;
1279 struct ib_cqe *wr_cqe;
1281 struct ib_sge *sg_list;
1283 enum ib_wr_opcode opcode;
1287 u32 invalidate_rkey;
1292 struct ib_send_wr wr;
1297 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1299 return container_of(wr, struct ib_rdma_wr, wr);
1302 struct ib_atomic_wr {
1303 struct ib_send_wr wr;
1307 u64 compare_add_mask;
1312 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1314 return container_of(wr, struct ib_atomic_wr, wr);
1318 struct ib_send_wr wr;
1325 u16 pkey_index; /* valid for GSI only */
1326 u8 port_num; /* valid for DR SMPs on switch only */
1329 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1331 return container_of(wr, struct ib_ud_wr, wr);
1335 struct ib_send_wr wr;
1341 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1343 return container_of(wr, struct ib_reg_wr, wr);
1347 struct ib_recv_wr *next;
1350 struct ib_cqe *wr_cqe;
1352 struct ib_sge *sg_list;
1356 enum ib_access_flags {
1357 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1358 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1359 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1360 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1361 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1362 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1363 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1364 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1366 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1370 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1371 * are hidden here instead of a uapi header!
1373 enum ib_mr_rereg_flags {
1374 IB_MR_REREG_TRANS = 1,
1375 IB_MR_REREG_PD = (1<<1),
1376 IB_MR_REREG_ACCESS = (1<<2),
1377 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1380 struct ib_fmr_attr {
1388 enum rdma_remove_reason {
1390 * Userspace requested uobject deletion or initial try
1391 * to remove uobject via cleanup. Call could fail
1393 RDMA_REMOVE_DESTROY,
1394 /* Context deletion. This call should delete the actual object itself */
1396 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1397 RDMA_REMOVE_DRIVER_REMOVE,
1398 /* uobj is being cleaned-up before being committed */
1402 struct ib_rdmacg_object {
1403 #ifdef CONFIG_CGROUP_RDMA
1404 struct rdma_cgroup *cg; /* owner rdma cgroup */
1408 struct ib_ucontext {
1409 struct ib_device *device;
1410 struct ib_uverbs_file *ufile;
1412 * 'closing' can be read by the driver only during a destroy callback,
1413 * it is set when we are closing the file descriptor and indicates
1414 * that mm_sem may be locked.
1418 bool cleanup_retryable;
1420 struct ib_rdmacg_object cg_obj;
1422 * Implementation details of the RDMA core, don't use in drivers:
1424 struct rdma_restrack_entry res;
1428 u64 user_handle; /* handle given to us by userspace */
1429 /* ufile & ucontext owning this object */
1430 struct ib_uverbs_file *ufile;
1431 /* FIXME, save memory: ufile->context == context */
1432 struct ib_ucontext *context; /* associated user context */
1433 void *object; /* containing object */
1434 struct list_head list; /* link to context's list */
1435 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1436 int id; /* index into kernel idr */
1438 atomic_t usecnt; /* protects exclusive access */
1439 struct rcu_head rcu; /* kfree_rcu() overhead */
1441 const struct uverbs_api_object *uapi_object;
1445 const void __user *inbuf;
1446 void __user *outbuf;
1454 struct ib_device *device;
1455 struct ib_uobject *uobject;
1456 atomic_t usecnt; /* count all resources */
1458 u32 unsafe_global_rkey;
1461 * Implementation details of the RDMA core, don't use in drivers:
1463 struct ib_mr *__internal_mr;
1464 struct rdma_restrack_entry res;
1468 struct ib_device *device;
1469 atomic_t usecnt; /* count all exposed resources */
1470 struct inode *inode;
1472 struct mutex tgt_qp_mutex;
1473 struct list_head tgt_qp_list;
1477 struct ib_device *device;
1479 struct ib_uobject *uobject;
1480 const struct ib_gid_attr *sgid_attr;
1481 enum rdma_ah_attr_type type;
1484 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1486 enum ib_poll_context {
1487 IB_POLL_DIRECT, /* caller context, no hw completions */
1488 IB_POLL_SOFTIRQ, /* poll from softirq context */
1489 IB_POLL_WORKQUEUE, /* poll from workqueue */
1490 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1494 struct ib_device *device;
1495 struct ib_uobject *uobject;
1496 ib_comp_handler comp_handler;
1497 void (*event_handler)(struct ib_event *, void *);
1500 atomic_t usecnt; /* count number of work queues */
1501 enum ib_poll_context poll_ctx;
1504 struct irq_poll iop;
1505 struct work_struct work;
1507 struct workqueue_struct *comp_wq;
1510 * Implementation details of the RDMA core, don't use in drivers:
1512 struct rdma_restrack_entry res;
1516 struct ib_device *device;
1518 struct ib_uobject *uobject;
1519 void (*event_handler)(struct ib_event *, void *);
1521 enum ib_srq_type srq_type;
1528 struct ib_xrcd *xrcd;
1535 enum ib_raw_packet_caps {
1536 /* Strip cvlan from incoming packet and report it in the matching work
1537 * completion is supported.
1539 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1540 /* Scatter FCS field of an incoming packet to host memory is supported.
1542 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1543 /* Checksum offloads are supported (for both send and receive). */
1544 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1545 /* When a packet is received for an RQ with no receive WQEs, the
1546 * packet processing is delayed.
1548 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1562 struct ib_device *device;
1563 struct ib_uobject *uobject;
1565 void (*event_handler)(struct ib_event *, void *);
1569 enum ib_wq_state state;
1570 enum ib_wq_type wq_type;
1575 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1576 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1577 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1578 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1581 struct ib_wq_init_attr {
1583 enum ib_wq_type wq_type;
1587 void (*event_handler)(struct ib_event *, void *);
1588 u32 create_flags; /* Use enum ib_wq_flags */
1591 enum ib_wq_attr_mask {
1592 IB_WQ_STATE = 1 << 0,
1593 IB_WQ_CUR_STATE = 1 << 1,
1594 IB_WQ_FLAGS = 1 << 2,
1598 enum ib_wq_state wq_state;
1599 enum ib_wq_state curr_wq_state;
1600 u32 flags; /* Use enum ib_wq_flags */
1601 u32 flags_mask; /* Use enum ib_wq_flags */
1604 struct ib_rwq_ind_table {
1605 struct ib_device *device;
1606 struct ib_uobject *uobject;
1609 u32 log_ind_tbl_size;
1610 struct ib_wq **ind_tbl;
1613 struct ib_rwq_ind_table_init_attr {
1614 u32 log_ind_tbl_size;
1615 /* Each entry is a pointer to Receive Work Queue */
1616 struct ib_wq **ind_tbl;
1619 enum port_pkey_state {
1620 IB_PORT_PKEY_NOT_VALID = 0,
1621 IB_PORT_PKEY_VALID = 1,
1622 IB_PORT_PKEY_LISTED = 2,
1625 struct ib_qp_security;
1627 struct ib_port_pkey {
1628 enum port_pkey_state state;
1631 struct list_head qp_list;
1632 struct list_head to_error_list;
1633 struct ib_qp_security *sec;
1636 struct ib_ports_pkeys {
1637 struct ib_port_pkey main;
1638 struct ib_port_pkey alt;
1641 struct ib_qp_security {
1643 struct ib_device *dev;
1644 /* Hold this mutex when changing port and pkey settings. */
1646 struct ib_ports_pkeys *ports_pkeys;
1647 /* A list of all open shared QP handles. Required to enforce security
1648 * properly for all users of a shared QP.
1650 struct list_head shared_qp_list;
1653 atomic_t error_list_count;
1654 struct completion error_complete;
1655 int error_comps_pending;
1659 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1660 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1663 struct ib_device *device;
1665 struct ib_cq *send_cq;
1666 struct ib_cq *recv_cq;
1669 struct list_head rdma_mrs;
1670 struct list_head sig_mrs;
1672 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1673 struct list_head xrcd_list;
1675 /* count times opened, mcast attaches, flow attaches */
1677 struct list_head open_list;
1678 struct ib_qp *real_qp;
1679 struct ib_uobject *uobject;
1680 void (*event_handler)(struct ib_event *, void *);
1682 /* sgid_attrs associated with the AV's */
1683 const struct ib_gid_attr *av_sgid_attr;
1684 const struct ib_gid_attr *alt_path_sgid_attr;
1688 enum ib_qp_type qp_type;
1689 struct ib_rwq_ind_table *rwq_ind_tbl;
1690 struct ib_qp_security *qp_sec;
1695 * Implementation details of the RDMA core, don't use in drivers:
1697 struct rdma_restrack_entry res;
1699 /* The counter the qp is bind to */
1700 struct rdma_counter *counter;
1704 struct ib_device *device;
1707 struct ib_uobject *uobject;
1712 struct ib_device *device;
1718 unsigned int page_size;
1719 enum ib_mr_type type;
1722 struct ib_uobject *uobject; /* user */
1723 struct list_head qp_entry; /* FR */
1727 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1729 * Implementation details of the RDMA core, don't use in drivers:
1731 struct rdma_restrack_entry res;
1735 struct ib_device *device;
1737 struct ib_uobject *uobject;
1739 enum ib_mw_type type;
1743 struct ib_device *device;
1745 struct list_head list;
1750 /* Supported steering options */
1751 enum ib_flow_attr_type {
1752 /* steering according to rule specifications */
1753 IB_FLOW_ATTR_NORMAL = 0x0,
1754 /* default unicast and multicast rule -
1755 * receive all Eth traffic which isn't steered to any QP
1757 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1758 /* default multicast rule -
1759 * receive all Eth multicast traffic which isn't steered to any QP
1761 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1762 /* sniffer rule - receive all port traffic */
1763 IB_FLOW_ATTR_SNIFFER = 0x3
1766 /* Supported steering header types */
1767 enum ib_flow_spec_type {
1769 IB_FLOW_SPEC_ETH = 0x20,
1770 IB_FLOW_SPEC_IB = 0x22,
1772 IB_FLOW_SPEC_IPV4 = 0x30,
1773 IB_FLOW_SPEC_IPV6 = 0x31,
1774 IB_FLOW_SPEC_ESP = 0x34,
1776 IB_FLOW_SPEC_TCP = 0x40,
1777 IB_FLOW_SPEC_UDP = 0x41,
1778 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1779 IB_FLOW_SPEC_GRE = 0x51,
1780 IB_FLOW_SPEC_MPLS = 0x60,
1781 IB_FLOW_SPEC_INNER = 0x100,
1783 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1784 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1785 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1786 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1788 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1789 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1791 /* Flow steering rule priority is set according to it's domain.
1792 * Lower domain value means higher priority.
1794 enum ib_flow_domain {
1795 IB_FLOW_DOMAIN_USER,
1796 IB_FLOW_DOMAIN_ETHTOOL,
1799 IB_FLOW_DOMAIN_NUM /* Must be last */
1802 enum ib_flow_flags {
1803 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1804 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1805 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1808 struct ib_flow_eth_filter {
1817 struct ib_flow_spec_eth {
1820 struct ib_flow_eth_filter val;
1821 struct ib_flow_eth_filter mask;
1824 struct ib_flow_ib_filter {
1831 struct ib_flow_spec_ib {
1834 struct ib_flow_ib_filter val;
1835 struct ib_flow_ib_filter mask;
1838 /* IPv4 header flags */
1839 enum ib_ipv4_flags {
1840 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1841 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1842 last have this flag set */
1845 struct ib_flow_ipv4_filter {
1856 struct ib_flow_spec_ipv4 {
1859 struct ib_flow_ipv4_filter val;
1860 struct ib_flow_ipv4_filter mask;
1863 struct ib_flow_ipv6_filter {
1874 struct ib_flow_spec_ipv6 {
1877 struct ib_flow_ipv6_filter val;
1878 struct ib_flow_ipv6_filter mask;
1881 struct ib_flow_tcp_udp_filter {
1888 struct ib_flow_spec_tcp_udp {
1891 struct ib_flow_tcp_udp_filter val;
1892 struct ib_flow_tcp_udp_filter mask;
1895 struct ib_flow_tunnel_filter {
1900 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1901 * the tunnel_id from val has the vni value
1903 struct ib_flow_spec_tunnel {
1906 struct ib_flow_tunnel_filter val;
1907 struct ib_flow_tunnel_filter mask;
1910 struct ib_flow_esp_filter {
1917 struct ib_flow_spec_esp {
1920 struct ib_flow_esp_filter val;
1921 struct ib_flow_esp_filter mask;
1924 struct ib_flow_gre_filter {
1925 __be16 c_ks_res0_ver;
1932 struct ib_flow_spec_gre {
1935 struct ib_flow_gre_filter val;
1936 struct ib_flow_gre_filter mask;
1939 struct ib_flow_mpls_filter {
1945 struct ib_flow_spec_mpls {
1948 struct ib_flow_mpls_filter val;
1949 struct ib_flow_mpls_filter mask;
1952 struct ib_flow_spec_action_tag {
1953 enum ib_flow_spec_type type;
1958 struct ib_flow_spec_action_drop {
1959 enum ib_flow_spec_type type;
1963 struct ib_flow_spec_action_handle {
1964 enum ib_flow_spec_type type;
1966 struct ib_flow_action *act;
1969 enum ib_counters_description {
1974 struct ib_flow_spec_action_count {
1975 enum ib_flow_spec_type type;
1977 struct ib_counters *counters;
1980 union ib_flow_spec {
1985 struct ib_flow_spec_eth eth;
1986 struct ib_flow_spec_ib ib;
1987 struct ib_flow_spec_ipv4 ipv4;
1988 struct ib_flow_spec_tcp_udp tcp_udp;
1989 struct ib_flow_spec_ipv6 ipv6;
1990 struct ib_flow_spec_tunnel tunnel;
1991 struct ib_flow_spec_esp esp;
1992 struct ib_flow_spec_gre gre;
1993 struct ib_flow_spec_mpls mpls;
1994 struct ib_flow_spec_action_tag flow_tag;
1995 struct ib_flow_spec_action_drop drop;
1996 struct ib_flow_spec_action_handle action;
1997 struct ib_flow_spec_action_count flow_count;
2000 struct ib_flow_attr {
2001 enum ib_flow_attr_type type;
2007 union ib_flow_spec flows[];
2012 struct ib_device *device;
2013 struct ib_uobject *uobject;
2016 enum ib_flow_action_type {
2017 IB_FLOW_ACTION_UNSPECIFIED,
2018 IB_FLOW_ACTION_ESP = 1,
2021 struct ib_flow_action_attrs_esp_keymats {
2022 enum ib_uverbs_flow_action_esp_keymat protocol;
2024 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2028 struct ib_flow_action_attrs_esp_replays {
2029 enum ib_uverbs_flow_action_esp_replay protocol;
2031 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2035 enum ib_flow_action_attrs_esp_flags {
2036 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2037 * This is done in order to share the same flags between user-space and
2038 * kernel and spare an unnecessary translation.
2042 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2043 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2046 struct ib_flow_spec_list {
2047 struct ib_flow_spec_list *next;
2048 union ib_flow_spec spec;
2051 struct ib_flow_action_attrs_esp {
2052 struct ib_flow_action_attrs_esp_keymats *keymat;
2053 struct ib_flow_action_attrs_esp_replays *replay;
2054 struct ib_flow_spec_list *encap;
2055 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2056 * Value of 0 is a valid value.
2062 /* Use enum ib_flow_action_attrs_esp_flags */
2064 u64 hard_limit_pkts;
2067 struct ib_flow_action {
2068 struct ib_device *device;
2069 struct ib_uobject *uobject;
2070 enum ib_flow_action_type type;
2077 enum ib_process_mad_flags {
2078 IB_MAD_IGNORE_MKEY = 1,
2079 IB_MAD_IGNORE_BKEY = 2,
2080 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2083 enum ib_mad_result {
2084 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2085 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2086 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2087 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2090 struct ib_port_cache {
2092 struct ib_pkey_cache *pkey;
2093 struct ib_gid_table *gid;
2095 enum ib_port_state port_state;
2100 struct ib_event_handler event_handler;
2103 struct ib_port_immutable {
2110 struct ib_port_data {
2111 struct ib_device *ib_dev;
2113 struct ib_port_immutable immutable;
2115 spinlock_t pkey_list_lock;
2116 struct list_head pkey_list;
2118 struct ib_port_cache cache;
2120 spinlock_t netdev_lock;
2121 struct net_device __rcu *netdev;
2122 struct hlist_node ndev_hash_link;
2123 struct rdma_port_counter port_counter;
2124 struct rdma_hw_stats *hw_stats;
2127 /* rdma netdev type - specifies protocol type */
2128 enum rdma_netdev_t {
2129 RDMA_NETDEV_OPA_VNIC,
2134 * struct rdma_netdev - rdma netdev
2135 * For cases where netstack interfacing is required.
2137 struct rdma_netdev {
2139 struct ib_device *hca;
2143 * cleanup function must be specified.
2144 * FIXME: This is only used for OPA_VNIC and that usage should be
2147 void (*free_rdma_netdev)(struct net_device *netdev);
2149 /* control functions */
2150 void (*set_id)(struct net_device *netdev, int id);
2152 int (*send)(struct net_device *dev, struct sk_buff *skb,
2153 struct ib_ah *address, u32 dqpn);
2155 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2156 union ib_gid *gid, u16 mlid,
2157 int set_qkey, u32 qkey);
2158 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2159 union ib_gid *gid, u16 mlid);
2162 struct rdma_netdev_alloc_params {
2168 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2169 struct net_device *netdev, void *param);
2172 struct ib_counters {
2173 struct ib_device *device;
2174 struct ib_uobject *uobject;
2175 /* num of objects attached */
2179 struct ib_counters_read_attr {
2182 u32 flags; /* use enum ib_read_counters_flags */
2185 struct uverbs_attr_bundle;
2187 struct iw_cm_conn_param;
2189 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2190 .size_##ib_struct = \
2191 (sizeof(struct drv_struct) + \
2192 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2193 BUILD_BUG_ON_ZERO( \
2194 !__same_type(((struct drv_struct *)NULL)->member, \
2197 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2198 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2200 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2201 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2203 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2206 * struct ib_device_ops - InfiniBand device operations
2207 * This structure defines all the InfiniBand device operations, providers will
2208 * need to define the supported operations, otherwise they will be set to null.
2210 struct ib_device_ops {
2211 struct module *owner;
2212 enum rdma_driver_id driver_id;
2214 unsigned int uverbs_no_driver_id_binding:1;
2216 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2217 const struct ib_send_wr **bad_send_wr);
2218 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2219 const struct ib_recv_wr **bad_recv_wr);
2220 void (*drain_rq)(struct ib_qp *qp);
2221 void (*drain_sq)(struct ib_qp *qp);
2222 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2223 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2224 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2225 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2226 int (*post_srq_recv)(struct ib_srq *srq,
2227 const struct ib_recv_wr *recv_wr,
2228 const struct ib_recv_wr **bad_recv_wr);
2229 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2230 u8 port_num, const struct ib_wc *in_wc,
2231 const struct ib_grh *in_grh,
2232 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2233 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2234 u16 *out_mad_pkey_index);
2235 int (*query_device)(struct ib_device *device,
2236 struct ib_device_attr *device_attr,
2237 struct ib_udata *udata);
2238 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2239 struct ib_device_modify *device_modify);
2240 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2241 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2243 int (*query_port)(struct ib_device *device, u8 port_num,
2244 struct ib_port_attr *port_attr);
2245 int (*modify_port)(struct ib_device *device, u8 port_num,
2246 int port_modify_mask,
2247 struct ib_port_modify *port_modify);
2249 * The following mandatory functions are used only at device
2250 * registration. Keep functions such as these at the end of this
2251 * structure to avoid cache line misses when accessing struct ib_device
2254 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2255 struct ib_port_immutable *immutable);
2256 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2259 * When calling get_netdev, the HW vendor's driver should return the
2260 * net device of device @device at port @port_num or NULL if such
2261 * a net device doesn't exist. The vendor driver should call dev_hold
2262 * on this net device. The HW vendor's device driver must guarantee
2263 * that this function returns NULL before the net device has finished
2264 * NETDEV_UNREGISTER state.
2266 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2268 * rdma netdev operation
2270 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2271 * must return -EOPNOTSUPP if it doesn't support the specified type.
2273 struct net_device *(*alloc_rdma_netdev)(
2274 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2275 const char *name, unsigned char name_assign_type,
2276 void (*setup)(struct net_device *));
2278 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2279 enum rdma_netdev_t type,
2280 struct rdma_netdev_alloc_params *params);
2282 * query_gid should be return GID value for @device, when @port_num
2283 * link layer is either IB or iWarp. It is no-op if @port_num port
2284 * is RoCE link layer.
2286 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2289 * When calling add_gid, the HW vendor's driver should add the gid
2290 * of device of port at gid index available at @attr. Meta-info of
2291 * that gid (for example, the network device related to this gid) is
2292 * available at @attr. @context allows the HW vendor driver to store
2293 * extra information together with a GID entry. The HW vendor driver may
2294 * allocate memory to contain this information and store it in @context
2295 * when a new GID entry is written to. Params are consistent until the
2296 * next call of add_gid or delete_gid. The function should return 0 on
2297 * success or error otherwise. The function could be called
2298 * concurrently for different ports. This function is only called when
2299 * roce_gid_table is used.
2301 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2303 * When calling del_gid, the HW vendor's driver should delete the
2304 * gid of device @device at gid index gid_index of port port_num
2305 * available in @attr.
2306 * Upon the deletion of a GID entry, the HW vendor must free any
2307 * allocated memory. The caller will clear @context afterwards.
2308 * This function is only called when roce_gid_table is used.
2310 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2311 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2313 int (*alloc_ucontext)(struct ib_ucontext *context,
2314 struct ib_udata *udata);
2315 void (*dealloc_ucontext)(struct ib_ucontext *context);
2316 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2317 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2318 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2319 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2320 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2321 u32 flags, struct ib_udata *udata);
2322 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2323 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2324 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2325 int (*create_srq)(struct ib_srq *srq,
2326 struct ib_srq_init_attr *srq_init_attr,
2327 struct ib_udata *udata);
2328 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2329 enum ib_srq_attr_mask srq_attr_mask,
2330 struct ib_udata *udata);
2331 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2332 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2333 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2334 struct ib_qp_init_attr *qp_init_attr,
2335 struct ib_udata *udata);
2336 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2337 int qp_attr_mask, struct ib_udata *udata);
2338 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2339 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2340 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2341 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2342 struct ib_udata *udata);
2343 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2344 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2345 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2346 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2347 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2348 u64 virt_addr, int mr_access_flags,
2349 struct ib_udata *udata);
2350 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2351 u64 virt_addr, int mr_access_flags,
2352 struct ib_pd *pd, struct ib_udata *udata);
2353 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2354 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2355 u32 max_num_sg, struct ib_udata *udata);
2356 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2357 u32 max_num_data_sg,
2358 u32 max_num_meta_sg);
2359 int (*advise_mr)(struct ib_pd *pd,
2360 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2361 struct ib_sge *sg_list, u32 num_sge,
2362 struct uverbs_attr_bundle *attrs);
2363 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2364 unsigned int *sg_offset);
2365 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2366 struct ib_mr_status *mr_status);
2367 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2368 struct ib_udata *udata);
2369 int (*dealloc_mw)(struct ib_mw *mw);
2370 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2371 struct ib_fmr_attr *fmr_attr);
2372 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2374 int (*unmap_fmr)(struct list_head *fmr_list);
2375 int (*dealloc_fmr)(struct ib_fmr *fmr);
2376 void (*invalidate_range)(struct ib_umem_odp *umem_odp,
2377 unsigned long start, unsigned long end);
2378 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2379 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2380 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2381 struct ib_udata *udata);
2382 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2383 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2384 struct ib_flow_attr *flow_attr,
2385 int domain, struct ib_udata *udata);
2386 int (*destroy_flow)(struct ib_flow *flow_id);
2387 struct ib_flow_action *(*create_flow_action_esp)(
2388 struct ib_device *device,
2389 const struct ib_flow_action_attrs_esp *attr,
2390 struct uverbs_attr_bundle *attrs);
2391 int (*destroy_flow_action)(struct ib_flow_action *action);
2392 int (*modify_flow_action_esp)(
2393 struct ib_flow_action *action,
2394 const struct ib_flow_action_attrs_esp *attr,
2395 struct uverbs_attr_bundle *attrs);
2396 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2398 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2399 struct ifla_vf_info *ivf);
2400 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2401 struct ifla_vf_stats *stats);
2402 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2404 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2405 struct ib_wq_init_attr *init_attr,
2406 struct ib_udata *udata);
2407 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2408 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2409 u32 wq_attr_mask, struct ib_udata *udata);
2410 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2411 struct ib_device *device,
2412 struct ib_rwq_ind_table_init_attr *init_attr,
2413 struct ib_udata *udata);
2414 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2415 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2416 struct ib_ucontext *context,
2417 struct ib_dm_alloc_attr *attr,
2418 struct uverbs_attr_bundle *attrs);
2419 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2420 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2421 struct ib_dm_mr_attr *attr,
2422 struct uverbs_attr_bundle *attrs);
2423 struct ib_counters *(*create_counters)(
2424 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2425 int (*destroy_counters)(struct ib_counters *counters);
2426 int (*read_counters)(struct ib_counters *counters,
2427 struct ib_counters_read_attr *counters_read_attr,
2428 struct uverbs_attr_bundle *attrs);
2429 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2430 int data_sg_nents, unsigned int *data_sg_offset,
2431 struct scatterlist *meta_sg, int meta_sg_nents,
2432 unsigned int *meta_sg_offset);
2435 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2436 * driver initialized data. The struct is kfree()'ed by the sysfs
2437 * core when the device is removed. A lifespan of -1 in the return
2438 * struct tells the core to set a default lifespan.
2440 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2443 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2444 * @index - The index in the value array we wish to have updated, or
2445 * num_counters if we want all stats updated
2447 * < 0 - Error, no counters updated
2448 * index - Updated the single counter pointed to by index
2449 * num_counters - Updated all counters (will reset the timestamp
2450 * and prevent further calls for lifespan milliseconds)
2451 * Drivers are allowed to update all counters in leiu of just the
2452 * one given in index at their option
2454 int (*get_hw_stats)(struct ib_device *device,
2455 struct rdma_hw_stats *stats, u8 port, int index);
2457 * This function is called once for each port when a ib device is
2460 int (*init_port)(struct ib_device *device, u8 port_num,
2461 struct kobject *port_sysfs);
2463 * Allows rdma drivers to add their own restrack attributes.
2465 int (*fill_res_entry)(struct sk_buff *msg,
2466 struct rdma_restrack_entry *entry);
2468 /* Device lifecycle callbacks */
2470 * Called after the device becomes registered, before clients are
2473 int (*enable_driver)(struct ib_device *dev);
2475 * This is called as part of ib_dealloc_device().
2477 void (*dealloc_driver)(struct ib_device *dev);
2479 /* iWarp CM callbacks */
2480 void (*iw_add_ref)(struct ib_qp *qp);
2481 void (*iw_rem_ref)(struct ib_qp *qp);
2482 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2483 int (*iw_connect)(struct iw_cm_id *cm_id,
2484 struct iw_cm_conn_param *conn_param);
2485 int (*iw_accept)(struct iw_cm_id *cm_id,
2486 struct iw_cm_conn_param *conn_param);
2487 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2489 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2490 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2492 * counter_bind_qp - Bind a QP to a counter.
2493 * @counter - The counter to be bound. If counter->id is zero then
2494 * the driver needs to allocate a new counter and set counter->id
2496 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2498 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2499 * counter and bind it onto the default one
2501 int (*counter_unbind_qp)(struct ib_qp *qp);
2503 * counter_dealloc -De-allocate the hw counter
2505 int (*counter_dealloc)(struct rdma_counter *counter);
2507 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2508 * the driver initialized data.
2510 struct rdma_hw_stats *(*counter_alloc_stats)(
2511 struct rdma_counter *counter);
2513 * counter_update_stats - Query the stats value of this counter
2515 int (*counter_update_stats)(struct rdma_counter *counter);
2517 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2518 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2519 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2520 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2521 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2524 struct ib_core_device {
2525 /* device must be the first element in structure until,
2526 * union of ib_core_device and device exists in ib_device.
2529 possible_net_t rdma_net;
2530 struct kobject *ports_kobj;
2531 struct list_head port_list;
2532 struct ib_device *owner; /* reach back to owner ib_device */
2535 struct rdma_restrack_root;
2537 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2538 struct device *dma_device;
2539 struct ib_device_ops ops;
2540 char name[IB_DEVICE_NAME_MAX];
2541 struct rcu_head rcu_head;
2543 struct list_head event_handler_list;
2544 spinlock_t event_handler_lock;
2546 struct rw_semaphore client_data_rwsem;
2547 struct xarray client_data;
2548 struct mutex unregistration_lock;
2550 struct ib_cache cache;
2552 * port_data is indexed by port number
2554 struct ib_port_data *port_data;
2556 int num_comp_vectors;
2560 struct ib_core_device coredev;
2563 /* First group for device attributes,
2564 * Second group for driver provided attributes (optional).
2565 * It is NULL terminated array.
2567 const struct attribute_group *groups[3];
2569 u64 uverbs_cmd_mask;
2570 u64 uverbs_ex_cmd_mask;
2572 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2576 /* Indicates kernel verbs support, should not be used in drivers */
2577 u16 kverbs_provider:1;
2578 /* CQ adaptive moderation (RDMA DIM) */
2582 struct ib_device_attr attrs;
2583 struct attribute_group *hw_stats_ag;
2584 struct rdma_hw_stats *hw_stats;
2586 #ifdef CONFIG_CGROUP_RDMA
2587 struct rdmacg_device cg_device;
2591 struct rdma_restrack_root *res;
2593 const struct uapi_definition *driver_def;
2596 * Positive refcount indicates that the device is currently
2597 * registered and cannot be unregistered.
2599 refcount_t refcount;
2600 struct completion unreg_completion;
2601 struct work_struct unregistration_work;
2603 const struct rdma_link_ops *link_ops;
2605 /* Protects compat_devs xarray modifications */
2606 struct mutex compat_devs_mutex;
2607 /* Maintains compat devices for each net namespace */
2608 struct xarray compat_devs;
2610 /* Used by iWarp CM */
2611 char iw_ifname[IFNAMSIZ];
2612 u32 iw_driver_flags;
2615 struct ib_client_nl_info;
2618 void (*add) (struct ib_device *);
2619 void (*remove)(struct ib_device *, void *client_data);
2620 void (*rename)(struct ib_device *dev, void *client_data);
2621 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2622 struct ib_client_nl_info *res);
2623 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2625 /* Returns the net_dev belonging to this ib_client and matching the
2627 * @dev: An RDMA device that the net_dev use for communication.
2628 * @port: A physical port number on the RDMA device.
2629 * @pkey: P_Key that the net_dev uses if applicable.
2630 * @gid: A GID that the net_dev uses to communicate.
2631 * @addr: An IP address the net_dev is configured with.
2632 * @client_data: The device's client data set by ib_set_client_data().
2634 * An ib_client that implements a net_dev on top of RDMA devices
2635 * (such as IP over IB) should implement this callback, allowing the
2636 * rdma_cm module to find the right net_dev for a given request.
2638 * The caller is responsible for calling dev_put on the returned
2640 struct net_device *(*get_net_dev_by_params)(
2641 struct ib_device *dev,
2644 const union ib_gid *gid,
2645 const struct sockaddr *addr,
2649 struct completion uses_zero;
2652 /* kverbs are not required by the client */
2657 * IB block DMA iterator
2659 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2660 * to a HW supported page size.
2662 struct ib_block_iter {
2663 /* internal states */
2664 struct scatterlist *__sg; /* sg holding the current aligned block */
2665 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2666 unsigned int __sg_nents; /* number of SG entries */
2667 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2668 unsigned int __pg_bit; /* alignment of current block */
2671 struct ib_device *_ib_alloc_device(size_t size);
2672 #define ib_alloc_device(drv_struct, member) \
2673 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2674 BUILD_BUG_ON_ZERO(offsetof( \
2675 struct drv_struct, member))), \
2676 struct drv_struct, member)
2678 void ib_dealloc_device(struct ib_device *device);
2680 void ib_get_device_fw_str(struct ib_device *device, char *str);
2682 int ib_register_device(struct ib_device *device, const char *name);
2683 void ib_unregister_device(struct ib_device *device);
2684 void ib_unregister_driver(enum rdma_driver_id driver_id);
2685 void ib_unregister_device_and_put(struct ib_device *device);
2686 void ib_unregister_device_queued(struct ib_device *ib_dev);
2688 int ib_register_client (struct ib_client *client);
2689 void ib_unregister_client(struct ib_client *client);
2691 void __rdma_block_iter_start(struct ib_block_iter *biter,
2692 struct scatterlist *sglist,
2694 unsigned long pgsz);
2695 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2698 * rdma_block_iter_dma_address - get the aligned dma address of the current
2699 * block held by the block iterator.
2700 * @biter: block iterator holding the memory block
2702 static inline dma_addr_t
2703 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2705 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2709 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2710 * @sglist: sglist to iterate over
2711 * @biter: block iterator holding the memory block
2712 * @nents: maximum number of sg entries to iterate over
2713 * @pgsz: best HW supported page size to use
2715 * Callers may use rdma_block_iter_dma_address() to get each
2716 * blocks aligned DMA address.
2718 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2719 for (__rdma_block_iter_start(biter, sglist, nents, \
2721 __rdma_block_iter_next(biter);)
2724 * ib_get_client_data - Get IB client context
2725 * @device:Device to get context for
2726 * @client:Client to get context for
2728 * ib_get_client_data() returns the client context data set with
2729 * ib_set_client_data(). This can only be called while the client is
2730 * registered to the device, once the ib_client remove() callback returns this
2733 static inline void *ib_get_client_data(struct ib_device *device,
2734 struct ib_client *client)
2736 return xa_load(&device->client_data, client->client_id);
2738 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2740 void ib_set_device_ops(struct ib_device *device,
2741 const struct ib_device_ops *ops);
2743 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2744 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2745 unsigned long pfn, unsigned long size, pgprot_t prot);
2747 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2748 struct vm_area_struct *vma,
2749 unsigned long pfn, unsigned long size,
2756 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2758 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2761 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2763 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2766 static inline bool ib_is_buffer_cleared(const void __user *p,
2772 if (len > USHRT_MAX)
2775 buf = memdup_user(p, len);
2779 ret = !memchr_inv(buf, 0, len);
2784 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2788 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2792 * ib_is_destroy_retryable - Check whether the uobject destruction
2794 * @ret: The initial destruction return code
2795 * @why: remove reason
2796 * @uobj: The uobject that is destroyed
2798 * This function is a helper function that IB layer and low-level drivers
2799 * can use to consider whether the destruction of the given uobject is
2801 * It checks the original return code, if it wasn't success the destruction
2802 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2803 * the remove reason. (i.e. why).
2804 * Must be called with the object locked for destroy.
2806 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2807 struct ib_uobject *uobj)
2809 return ret && (why == RDMA_REMOVE_DESTROY ||
2810 uobj->context->cleanup_retryable);
2814 * ib_destroy_usecnt - Called during destruction to check the usecnt
2815 * @usecnt: The usecnt atomic
2816 * @why: remove reason
2817 * @uobj: The uobject that is destroyed
2819 * Non-zero usecnts will block destruction unless destruction was triggered by
2820 * a ucontext cleanup.
2822 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2823 enum rdma_remove_reason why,
2824 struct ib_uobject *uobj)
2826 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2832 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2833 * contains all required attributes and no attributes not allowed for
2834 * the given QP state transition.
2835 * @cur_state: Current QP state
2836 * @next_state: Next QP state
2838 * @mask: Mask of supplied QP attributes
2840 * This function is a helper function that a low-level driver's
2841 * modify_qp method can use to validate the consumer's input. It
2842 * checks that cur_state and next_state are valid QP states, that a
2843 * transition from cur_state to next_state is allowed by the IB spec,
2844 * and that the attribute mask supplied is allowed for the transition.
2846 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2847 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2849 void ib_register_event_handler(struct ib_event_handler *event_handler);
2850 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2851 void ib_dispatch_event(struct ib_event *event);
2853 int ib_query_port(struct ib_device *device,
2854 u8 port_num, struct ib_port_attr *port_attr);
2856 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2860 * rdma_cap_ib_switch - Check if the device is IB switch
2861 * @device: Device to check
2863 * Device driver is responsible for setting is_switch bit on
2864 * in ib_device structure at init time.
2866 * Return: true if the device is IB switch.
2868 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2870 return device->is_switch;
2874 * rdma_start_port - Return the first valid port number for the device
2877 * @device: Device to be checked
2879 * Return start port number
2881 static inline u8 rdma_start_port(const struct ib_device *device)
2883 return rdma_cap_ib_switch(device) ? 0 : 1;
2887 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2888 * @device - The struct ib_device * to iterate over
2889 * @iter - The unsigned int to store the port number
2891 #define rdma_for_each_port(device, iter) \
2892 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2893 unsigned int, iter))); \
2894 iter <= rdma_end_port(device); (iter)++)
2897 * rdma_end_port - Return the last valid port number for the device
2900 * @device: Device to be checked
2902 * Return last port number
2904 static inline u8 rdma_end_port(const struct ib_device *device)
2906 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2909 static inline int rdma_is_port_valid(const struct ib_device *device,
2912 return (port >= rdma_start_port(device) &&
2913 port <= rdma_end_port(device));
2916 static inline bool rdma_is_grh_required(const struct ib_device *device,
2919 return device->port_data[port_num].immutable.core_cap_flags &
2920 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2923 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2925 return device->port_data[port_num].immutable.core_cap_flags &
2926 RDMA_CORE_CAP_PROT_IB;
2929 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2931 return device->port_data[port_num].immutable.core_cap_flags &
2932 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2935 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2937 return device->port_data[port_num].immutable.core_cap_flags &
2938 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2941 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2943 return device->port_data[port_num].immutable.core_cap_flags &
2944 RDMA_CORE_CAP_PROT_ROCE;
2947 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2949 return device->port_data[port_num].immutable.core_cap_flags &
2950 RDMA_CORE_CAP_PROT_IWARP;
2953 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2955 return rdma_protocol_ib(device, port_num) ||
2956 rdma_protocol_roce(device, port_num);
2959 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2961 return device->port_data[port_num].immutable.core_cap_flags &
2962 RDMA_CORE_CAP_PROT_RAW_PACKET;
2965 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2967 return device->port_data[port_num].immutable.core_cap_flags &
2968 RDMA_CORE_CAP_PROT_USNIC;
2972 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2973 * Management Datagrams.
2974 * @device: Device to check
2975 * @port_num: Port number to check
2977 * Management Datagrams (MAD) are a required part of the InfiniBand
2978 * specification and are supported on all InfiniBand devices. A slightly
2979 * extended version are also supported on OPA interfaces.
2981 * Return: true if the port supports sending/receiving of MAD packets.
2983 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2985 return device->port_data[port_num].immutable.core_cap_flags &
2986 RDMA_CORE_CAP_IB_MAD;
2990 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2991 * Management Datagrams.
2992 * @device: Device to check
2993 * @port_num: Port number to check
2995 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2996 * datagrams with their own versions. These OPA MADs share many but not all of
2997 * the characteristics of InfiniBand MADs.
2999 * OPA MADs differ in the following ways:
3001 * 1) MADs are variable size up to 2K
3002 * IBTA defined MADs remain fixed at 256 bytes
3003 * 2) OPA SMPs must carry valid PKeys
3004 * 3) OPA SMP packets are a different format
3006 * Return: true if the port supports OPA MAD packet formats.
3008 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3010 return device->port_data[port_num].immutable.core_cap_flags &
3011 RDMA_CORE_CAP_OPA_MAD;
3015 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3016 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3017 * @device: Device to check
3018 * @port_num: Port number to check
3020 * Each InfiniBand node is required to provide a Subnet Management Agent
3021 * that the subnet manager can access. Prior to the fabric being fully
3022 * configured by the subnet manager, the SMA is accessed via a well known
3023 * interface called the Subnet Management Interface (SMI). This interface
3024 * uses directed route packets to communicate with the SM to get around the
3025 * chicken and egg problem of the SM needing to know what's on the fabric
3026 * in order to configure the fabric, and needing to configure the fabric in
3027 * order to send packets to the devices on the fabric. These directed
3028 * route packets do not need the fabric fully configured in order to reach
3029 * their destination. The SMI is the only method allowed to send
3030 * directed route packets on an InfiniBand fabric.
3032 * Return: true if the port provides an SMI.
3034 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3036 return device->port_data[port_num].immutable.core_cap_flags &
3037 RDMA_CORE_CAP_IB_SMI;
3041 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3042 * Communication Manager.
3043 * @device: Device to check
3044 * @port_num: Port number to check
3046 * The InfiniBand Communication Manager is one of many pre-defined General
3047 * Service Agents (GSA) that are accessed via the General Service
3048 * Interface (GSI). It's role is to facilitate establishment of connections
3049 * between nodes as well as other management related tasks for established
3052 * Return: true if the port supports an IB CM (this does not guarantee that
3053 * a CM is actually running however).
3055 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3057 return device->port_data[port_num].immutable.core_cap_flags &
3058 RDMA_CORE_CAP_IB_CM;
3062 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3063 * Communication Manager.
3064 * @device: Device to check
3065 * @port_num: Port number to check
3067 * Similar to above, but specific to iWARP connections which have a different
3068 * managment protocol than InfiniBand.
3070 * Return: true if the port supports an iWARP CM (this does not guarantee that
3071 * a CM is actually running however).
3073 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3075 return device->port_data[port_num].immutable.core_cap_flags &
3076 RDMA_CORE_CAP_IW_CM;
3080 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3081 * Subnet Administration.
3082 * @device: Device to check
3083 * @port_num: Port number to check
3085 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3086 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3087 * fabrics, devices should resolve routes to other hosts by contacting the
3088 * SA to query the proper route.
3090 * Return: true if the port should act as a client to the fabric Subnet
3091 * Administration interface. This does not imply that the SA service is
3094 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3096 return device->port_data[port_num].immutable.core_cap_flags &
3097 RDMA_CORE_CAP_IB_SA;
3101 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3103 * @device: Device to check
3104 * @port_num: Port number to check
3106 * InfiniBand multicast registration is more complex than normal IPv4 or
3107 * IPv6 multicast registration. Each Host Channel Adapter must register
3108 * with the Subnet Manager when it wishes to join a multicast group. It
3109 * should do so only once regardless of how many queue pairs it subscribes
3110 * to this group. And it should leave the group only after all queue pairs
3111 * attached to the group have been detached.
3113 * Return: true if the port must undertake the additional adminstrative
3114 * overhead of registering/unregistering with the SM and tracking of the
3115 * total number of queue pairs attached to the multicast group.
3117 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3119 return rdma_cap_ib_sa(device, port_num);
3123 * rdma_cap_af_ib - Check if the port of device has the capability
3124 * Native Infiniband Address.
3125 * @device: Device to check
3126 * @port_num: Port number to check
3128 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3129 * GID. RoCE uses a different mechanism, but still generates a GID via
3130 * a prescribed mechanism and port specific data.
3132 * Return: true if the port uses a GID address to identify devices on the
3135 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3137 return device->port_data[port_num].immutable.core_cap_flags &
3138 RDMA_CORE_CAP_AF_IB;
3142 * rdma_cap_eth_ah - Check if the port of device has the capability
3143 * Ethernet Address Handle.
3144 * @device: Device to check
3145 * @port_num: Port number to check
3147 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3148 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3149 * port. Normally, packet headers are generated by the sending host
3150 * adapter, but when sending connectionless datagrams, we must manually
3151 * inject the proper headers for the fabric we are communicating over.
3153 * Return: true if we are running as a RoCE port and must force the
3154 * addition of a Global Route Header built from our Ethernet Address
3155 * Handle into our header list for connectionless packets.
3157 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3159 return device->port_data[port_num].immutable.core_cap_flags &
3160 RDMA_CORE_CAP_ETH_AH;
3164 * rdma_cap_opa_ah - Check if the port of device supports
3165 * OPA Address handles
3166 * @device: Device to check
3167 * @port_num: Port number to check
3169 * Return: true if we are running on an OPA device which supports
3170 * the extended OPA addressing.
3172 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3174 return (device->port_data[port_num].immutable.core_cap_flags &
3175 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3179 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3182 * @port_num: Port number
3184 * This MAD size includes the MAD headers and MAD payload. No other headers
3187 * Return the max MAD size required by the Port. Will return 0 if the port
3188 * does not support MADs
3190 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3192 return device->port_data[port_num].immutable.max_mad_size;
3196 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3197 * @device: Device to check
3198 * @port_num: Port number to check
3200 * RoCE GID table mechanism manages the various GIDs for a device.
3202 * NOTE: if allocating the port's GID table has failed, this call will still
3203 * return true, but any RoCE GID table API will fail.
3205 * Return: true if the port uses RoCE GID table mechanism in order to manage
3208 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3211 return rdma_protocol_roce(device, port_num) &&
3212 device->ops.add_gid && device->ops.del_gid;
3216 * Check if the device supports READ W/ INVALIDATE.
3218 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3221 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3222 * has support for it yet.
3224 return rdma_protocol_iwarp(dev, port_num);
3228 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3231 * @pgsz_bitmap: bitmap of HW supported page sizes
3233 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3234 unsigned long pgsz_bitmap)
3236 unsigned long align;
3239 align = addr & -addr;
3241 /* Find page bit such that addr is aligned to the highest supported
3244 pgsz = pgsz_bitmap & ~(-align << 1);
3246 return __ffs(pgsz_bitmap);
3251 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3253 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3254 struct ifla_vf_info *info);
3255 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3256 struct ifla_vf_stats *stats);
3257 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3260 int ib_query_pkey(struct ib_device *device,
3261 u8 port_num, u16 index, u16 *pkey);
3263 int ib_modify_device(struct ib_device *device,
3264 int device_modify_mask,
3265 struct ib_device_modify *device_modify);
3267 int ib_modify_port(struct ib_device *device,
3268 u8 port_num, int port_modify_mask,
3269 struct ib_port_modify *port_modify);
3271 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3272 u8 *port_num, u16 *index);
3274 int ib_find_pkey(struct ib_device *device,
3275 u8 port_num, u16 pkey, u16 *index);
3279 * Create a memory registration for all memory in the system and place
3280 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3281 * ULPs to avoid the overhead of dynamic MRs.
3283 * This flag is generally considered unsafe and must only be used in
3284 * extremly trusted environments. Every use of it will log a warning
3285 * in the kernel log.
3287 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3290 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3291 const char *caller);
3293 #define ib_alloc_pd(device, flags) \
3294 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3297 * ib_dealloc_pd_user - Deallocate kernel/user PD
3298 * @pd: The protection domain
3299 * @udata: Valid user data or NULL for kernel objects
3301 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3304 * ib_dealloc_pd - Deallocate kernel PD
3305 * @pd: The protection domain
3307 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3309 static inline void ib_dealloc_pd(struct ib_pd *pd)
3311 ib_dealloc_pd_user(pd, NULL);
3314 enum rdma_create_ah_flags {
3315 /* In a sleepable context */
3316 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3320 * rdma_create_ah - Creates an address handle for the given address vector.
3321 * @pd: The protection domain associated with the address handle.
3322 * @ah_attr: The attributes of the address vector.
3323 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3325 * The address handle is used to reference a local or global destination
3326 * in all UD QP post sends.
3328 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3332 * rdma_create_user_ah - Creates an address handle for the given address vector.
3333 * It resolves destination mac address for ah attribute of RoCE type.
3334 * @pd: The protection domain associated with the address handle.
3335 * @ah_attr: The attributes of the address vector.
3336 * @udata: pointer to user's input output buffer information need by
3339 * It returns 0 on success and returns appropriate error code on error.
3340 * The address handle is used to reference a local or global destination
3341 * in all UD QP post sends.
3343 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3344 struct rdma_ah_attr *ah_attr,
3345 struct ib_udata *udata);
3347 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3349 * @hdr: the L3 header to parse
3350 * @net_type: type of header to parse
3351 * @sgid: place to store source gid
3352 * @dgid: place to store destination gid
3354 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3355 enum rdma_network_type net_type,
3356 union ib_gid *sgid, union ib_gid *dgid);
3359 * ib_get_rdma_header_version - Get the header version
3360 * @hdr: the L3 header to parse
3362 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3365 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3367 * @device: Device on which the received message arrived.
3368 * @port_num: Port on which the received message arrived.
3369 * @wc: Work completion associated with the received message.
3370 * @grh: References the received global route header. This parameter is
3371 * ignored unless the work completion indicates that the GRH is valid.
3372 * @ah_attr: Returned attributes that can be used when creating an address
3373 * handle for replying to the message.
3374 * When ib_init_ah_attr_from_wc() returns success,
3375 * (a) for IB link layer it optionally contains a reference to SGID attribute
3376 * when GRH is present for IB link layer.
3377 * (b) for RoCE link layer it contains a reference to SGID attribute.
3378 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3379 * attributes which are initialized using ib_init_ah_attr_from_wc().
3382 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3383 const struct ib_wc *wc, const struct ib_grh *grh,
3384 struct rdma_ah_attr *ah_attr);
3387 * ib_create_ah_from_wc - Creates an address handle associated with the
3388 * sender of the specified work completion.
3389 * @pd: The protection domain associated with the address handle.
3390 * @wc: Work completion information associated with a received message.
3391 * @grh: References the received global route header. This parameter is
3392 * ignored unless the work completion indicates that the GRH is valid.
3393 * @port_num: The outbound port number to associate with the address.
3395 * The address handle is used to reference a local or global destination
3396 * in all UD QP post sends.
3398 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3399 const struct ib_grh *grh, u8 port_num);
3402 * rdma_modify_ah - Modifies the address vector associated with an address
3404 * @ah: The address handle to modify.
3405 * @ah_attr: The new address vector attributes to associate with the
3408 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3411 * rdma_query_ah - Queries the address vector associated with an address
3413 * @ah: The address handle to query.
3414 * @ah_attr: The address vector attributes associated with the address
3417 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3419 enum rdma_destroy_ah_flags {
3420 /* In a sleepable context */
3421 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3425 * rdma_destroy_ah_user - Destroys an address handle.
3426 * @ah: The address handle to destroy.
3427 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3428 * @udata: Valid user data or NULL for kernel objects
3430 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3433 * rdma_destroy_ah - Destroys an kernel address handle.
3434 * @ah: The address handle to destroy.
3435 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3437 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3439 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3441 return rdma_destroy_ah_user(ah, flags, NULL);
3445 * ib_create_srq - Creates a SRQ associated with the specified protection
3447 * @pd: The protection domain associated with the SRQ.
3448 * @srq_init_attr: A list of initial attributes required to create the
3449 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3450 * the actual capabilities of the created SRQ.
3452 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3453 * requested size of the SRQ, and set to the actual values allocated
3454 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3455 * will always be at least as large as the requested values.
3457 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3458 struct ib_srq_init_attr *srq_init_attr);
3461 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3462 * @srq: The SRQ to modify.
3463 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3464 * the current values of selected SRQ attributes are returned.
3465 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3466 * are being modified.
3468 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3469 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3470 * the number of receives queued drops below the limit.
3472 int ib_modify_srq(struct ib_srq *srq,
3473 struct ib_srq_attr *srq_attr,
3474 enum ib_srq_attr_mask srq_attr_mask);
3477 * ib_query_srq - Returns the attribute list and current values for the
3479 * @srq: The SRQ to query.
3480 * @srq_attr: The attributes of the specified SRQ.
3482 int ib_query_srq(struct ib_srq *srq,
3483 struct ib_srq_attr *srq_attr);
3486 * ib_destroy_srq_user - Destroys the specified SRQ.
3487 * @srq: The SRQ to destroy.
3488 * @udata: Valid user data or NULL for kernel objects
3490 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3493 * ib_destroy_srq - Destroys the specified kernel SRQ.
3494 * @srq: The SRQ to destroy.
3496 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3498 static inline int ib_destroy_srq(struct ib_srq *srq)
3500 return ib_destroy_srq_user(srq, NULL);
3504 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3505 * @srq: The SRQ to post the work request on.
3506 * @recv_wr: A list of work requests to post on the receive queue.
3507 * @bad_recv_wr: On an immediate failure, this parameter will reference
3508 * the work request that failed to be posted on the QP.
3510 static inline int ib_post_srq_recv(struct ib_srq *srq,
3511 const struct ib_recv_wr *recv_wr,
3512 const struct ib_recv_wr **bad_recv_wr)
3514 const struct ib_recv_wr *dummy;
3516 return srq->device->ops.post_srq_recv(srq, recv_wr,
3517 bad_recv_wr ? : &dummy);
3521 * ib_create_qp_user - Creates a QP associated with the specified protection
3523 * @pd: The protection domain associated with the QP.
3524 * @qp_init_attr: A list of initial attributes required to create the
3525 * QP. If QP creation succeeds, then the attributes are updated to
3526 * the actual capabilities of the created QP.
3527 * @udata: Valid user data or NULL for kernel objects
3529 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3530 struct ib_qp_init_attr *qp_init_attr,
3531 struct ib_udata *udata);
3534 * ib_create_qp - Creates a kernel QP associated with the specified protection
3536 * @pd: The protection domain associated with the QP.
3537 * @qp_init_attr: A list of initial attributes required to create the
3538 * QP. If QP creation succeeds, then the attributes are updated to
3539 * the actual capabilities of the created QP.
3540 * @udata: Valid user data or NULL for kernel objects
3542 * NOTE: for user qp use ib_create_qp_user with valid udata!
3544 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3545 struct ib_qp_init_attr *qp_init_attr)
3547 return ib_create_qp_user(pd, qp_init_attr, NULL);
3551 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3552 * @qp: The QP to modify.
3553 * @attr: On input, specifies the QP attributes to modify. On output,
3554 * the current values of selected QP attributes are returned.
3555 * @attr_mask: A bit-mask used to specify which attributes of the QP
3556 * are being modified.
3557 * @udata: pointer to user's input output buffer information
3558 * are being modified.
3559 * It returns 0 on success and returns appropriate error code on error.
3561 int ib_modify_qp_with_udata(struct ib_qp *qp,
3562 struct ib_qp_attr *attr,
3564 struct ib_udata *udata);
3567 * ib_modify_qp - Modifies the attributes for the specified QP and then
3568 * transitions the QP to the given state.
3569 * @qp: The QP to modify.
3570 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3571 * the current values of selected QP attributes are returned.
3572 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3573 * are being modified.
3575 int ib_modify_qp(struct ib_qp *qp,
3576 struct ib_qp_attr *qp_attr,
3580 * ib_query_qp - Returns the attribute list and current values for the
3582 * @qp: The QP to query.
3583 * @qp_attr: The attributes of the specified QP.
3584 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3585 * @qp_init_attr: Additional attributes of the selected QP.
3587 * The qp_attr_mask may be used to limit the query to gathering only the
3588 * selected attributes.
3590 int ib_query_qp(struct ib_qp *qp,
3591 struct ib_qp_attr *qp_attr,
3593 struct ib_qp_init_attr *qp_init_attr);
3596 * ib_destroy_qp - Destroys the specified QP.
3597 * @qp: The QP to destroy.
3598 * @udata: Valid udata or NULL for kernel objects
3600 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3603 * ib_destroy_qp - Destroys the specified kernel QP.
3604 * @qp: The QP to destroy.
3606 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3608 static inline int ib_destroy_qp(struct ib_qp *qp)
3610 return ib_destroy_qp_user(qp, NULL);
3614 * ib_open_qp - Obtain a reference to an existing sharable QP.
3615 * @xrcd - XRC domain
3616 * @qp_open_attr: Attributes identifying the QP to open.
3618 * Returns a reference to a sharable QP.
3620 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3621 struct ib_qp_open_attr *qp_open_attr);
3624 * ib_close_qp - Release an external reference to a QP.
3625 * @qp: The QP handle to release
3627 * The opened QP handle is released by the caller. The underlying
3628 * shared QP is not destroyed until all internal references are released.
3630 int ib_close_qp(struct ib_qp *qp);
3633 * ib_post_send - Posts a list of work requests to the send queue of
3635 * @qp: The QP to post the work request on.
3636 * @send_wr: A list of work requests to post on the send queue.
3637 * @bad_send_wr: On an immediate failure, this parameter will reference
3638 * the work request that failed to be posted on the QP.
3640 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3641 * error is returned, the QP state shall not be affected,
3642 * ib_post_send() will return an immediate error after queueing any
3643 * earlier work requests in the list.
3645 static inline int ib_post_send(struct ib_qp *qp,
3646 const struct ib_send_wr *send_wr,
3647 const struct ib_send_wr **bad_send_wr)
3649 const struct ib_send_wr *dummy;
3651 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3655 * ib_post_recv - Posts a list of work requests to the receive queue of
3657 * @qp: The QP to post the work request on.
3658 * @recv_wr: A list of work requests to post on the receive queue.
3659 * @bad_recv_wr: On an immediate failure, this parameter will reference
3660 * the work request that failed to be posted on the QP.
3662 static inline int ib_post_recv(struct ib_qp *qp,
3663 const struct ib_recv_wr *recv_wr,
3664 const struct ib_recv_wr **bad_recv_wr)
3666 const struct ib_recv_wr *dummy;
3668 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3671 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3672 int nr_cqe, int comp_vector,
3673 enum ib_poll_context poll_ctx,
3674 const char *caller, struct ib_udata *udata);
3677 * ib_alloc_cq_user: Allocate kernel/user CQ
3678 * @dev: The IB device
3679 * @private: Private data attached to the CQE
3680 * @nr_cqe: Number of CQEs in the CQ
3681 * @comp_vector: Completion vector used for the IRQs
3682 * @poll_ctx: Context used for polling the CQ
3683 * @udata: Valid user data or NULL for kernel objects
3685 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3686 void *private, int nr_cqe,
3688 enum ib_poll_context poll_ctx,
3689 struct ib_udata *udata)
3691 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3692 KBUILD_MODNAME, udata);
3696 * ib_alloc_cq: Allocate kernel CQ
3697 * @dev: The IB device
3698 * @private: Private data attached to the CQE
3699 * @nr_cqe: Number of CQEs in the CQ
3700 * @comp_vector: Completion vector used for the IRQs
3701 * @poll_ctx: Context used for polling the CQ
3703 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3705 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3706 int nr_cqe, int comp_vector,
3707 enum ib_poll_context poll_ctx)
3709 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3714 * ib_free_cq_user - Free kernel/user CQ
3715 * @cq: The CQ to free
3716 * @udata: Valid user data or NULL for kernel objects
3718 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3721 * ib_free_cq - Free kernel CQ
3722 * @cq: The CQ to free
3724 * NOTE: for user cq use ib_free_cq_user with valid udata!
3726 static inline void ib_free_cq(struct ib_cq *cq)
3728 ib_free_cq_user(cq, NULL);
3731 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3734 * ib_create_cq - Creates a CQ on the specified device.
3735 * @device: The device on which to create the CQ.
3736 * @comp_handler: A user-specified callback that is invoked when a
3737 * completion event occurs on the CQ.
3738 * @event_handler: A user-specified callback that is invoked when an
3739 * asynchronous event not associated with a completion occurs on the CQ.
3740 * @cq_context: Context associated with the CQ returned to the user via
3741 * the associated completion and event handlers.
3742 * @cq_attr: The attributes the CQ should be created upon.
3744 * Users can examine the cq structure to determine the actual CQ size.
3746 struct ib_cq *__ib_create_cq(struct ib_device *device,
3747 ib_comp_handler comp_handler,
3748 void (*event_handler)(struct ib_event *, void *),
3750 const struct ib_cq_init_attr *cq_attr,
3751 const char *caller);
3752 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3753 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3756 * ib_resize_cq - Modifies the capacity of the CQ.
3757 * @cq: The CQ to resize.
3758 * @cqe: The minimum size of the CQ.
3760 * Users can examine the cq structure to determine the actual CQ size.
3762 int ib_resize_cq(struct ib_cq *cq, int cqe);
3765 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3766 * @cq: The CQ to modify.
3767 * @cq_count: number of CQEs that will trigger an event
3768 * @cq_period: max period of time in usec before triggering an event
3771 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3774 * ib_destroy_cq_user - Destroys the specified CQ.
3775 * @cq: The CQ to destroy.
3776 * @udata: Valid user data or NULL for kernel objects
3778 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3781 * ib_destroy_cq - Destroys the specified kernel CQ.
3782 * @cq: The CQ to destroy.
3784 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3786 static inline void ib_destroy_cq(struct ib_cq *cq)
3788 ib_destroy_cq_user(cq, NULL);
3792 * ib_poll_cq - poll a CQ for completion(s)
3793 * @cq:the CQ being polled
3794 * @num_entries:maximum number of completions to return
3795 * @wc:array of at least @num_entries &struct ib_wc where completions
3798 * Poll a CQ for (possibly multiple) completions. If the return value
3799 * is < 0, an error occurred. If the return value is >= 0, it is the
3800 * number of completions returned. If the return value is
3801 * non-negative and < num_entries, then the CQ was emptied.
3803 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3806 return cq->device->ops.poll_cq(cq, num_entries, wc);
3810 * ib_req_notify_cq - Request completion notification on a CQ.
3811 * @cq: The CQ to generate an event for.
3813 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3814 * to request an event on the next solicited event or next work
3815 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3816 * may also be |ed in to request a hint about missed events, as
3820 * < 0 means an error occurred while requesting notification
3821 * == 0 means notification was requested successfully, and if
3822 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3823 * were missed and it is safe to wait for another event. In
3824 * this case is it guaranteed that any work completions added
3825 * to the CQ since the last CQ poll will trigger a completion
3826 * notification event.
3827 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3828 * in. It means that the consumer must poll the CQ again to
3829 * make sure it is empty to avoid missing an event because of a
3830 * race between requesting notification and an entry being
3831 * added to the CQ. This return value means it is possible
3832 * (but not guaranteed) that a work completion has been added
3833 * to the CQ since the last poll without triggering a
3834 * completion notification event.
3836 static inline int ib_req_notify_cq(struct ib_cq *cq,
3837 enum ib_cq_notify_flags flags)
3839 return cq->device->ops.req_notify_cq(cq, flags);
3843 * ib_req_ncomp_notif - Request completion notification when there are
3844 * at least the specified number of unreaped completions on the CQ.
3845 * @cq: The CQ to generate an event for.
3846 * @wc_cnt: The number of unreaped completions that should be on the
3847 * CQ before an event is generated.
3849 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3851 return cq->device->ops.req_ncomp_notif ?
3852 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3857 * ib_dma_mapping_error - check a DMA addr for error
3858 * @dev: The device for which the dma_addr was created
3859 * @dma_addr: The DMA address to check
3861 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3863 return dma_mapping_error(dev->dma_device, dma_addr);
3867 * ib_dma_map_single - Map a kernel virtual address to DMA address
3868 * @dev: The device for which the dma_addr is to be created
3869 * @cpu_addr: The kernel virtual address
3870 * @size: The size of the region in bytes
3871 * @direction: The direction of the DMA
3873 static inline u64 ib_dma_map_single(struct ib_device *dev,
3874 void *cpu_addr, size_t size,
3875 enum dma_data_direction direction)
3877 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3881 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3882 * @dev: The device for which the DMA address was created
3883 * @addr: The DMA address
3884 * @size: The size of the region in bytes
3885 * @direction: The direction of the DMA
3887 static inline void ib_dma_unmap_single(struct ib_device *dev,
3888 u64 addr, size_t size,
3889 enum dma_data_direction direction)
3891 dma_unmap_single(dev->dma_device, addr, size, direction);
3895 * ib_dma_map_page - Map a physical page to DMA address
3896 * @dev: The device for which the dma_addr is to be created
3897 * @page: The page to be mapped
3898 * @offset: The offset within the page
3899 * @size: The size of the region in bytes
3900 * @direction: The direction of the DMA
3902 static inline u64 ib_dma_map_page(struct ib_device *dev,
3904 unsigned long offset,
3906 enum dma_data_direction direction)
3908 return dma_map_page(dev->dma_device, page, offset, size, direction);
3912 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3913 * @dev: The device for which the DMA address was created
3914 * @addr: The DMA address
3915 * @size: The size of the region in bytes
3916 * @direction: The direction of the DMA
3918 static inline void ib_dma_unmap_page(struct ib_device *dev,
3919 u64 addr, size_t size,
3920 enum dma_data_direction direction)
3922 dma_unmap_page(dev->dma_device, addr, size, direction);
3926 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3927 * @dev: The device for which the DMA addresses are to be created
3928 * @sg: The array of scatter/gather entries
3929 * @nents: The number of scatter/gather entries
3930 * @direction: The direction of the DMA
3932 static inline int ib_dma_map_sg(struct ib_device *dev,
3933 struct scatterlist *sg, int nents,
3934 enum dma_data_direction direction)
3936 return dma_map_sg(dev->dma_device, sg, nents, direction);
3940 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3941 * @dev: The device for which the DMA addresses were created
3942 * @sg: The array of scatter/gather entries
3943 * @nents: The number of scatter/gather entries
3944 * @direction: The direction of the DMA
3946 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3947 struct scatterlist *sg, int nents,
3948 enum dma_data_direction direction)
3950 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3953 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3954 struct scatterlist *sg, int nents,
3955 enum dma_data_direction direction,
3956 unsigned long dma_attrs)
3958 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3962 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3963 struct scatterlist *sg, int nents,
3964 enum dma_data_direction direction,
3965 unsigned long dma_attrs)
3967 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3971 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3972 * @dev: The device to query
3974 * The returned value represents a size in bytes.
3976 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3978 struct device_dma_parameters *p = dev->dma_device->dma_parms;
3980 return p ? p->max_segment_size : UINT_MAX;
3984 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3985 * @dev: The device for which the DMA address was created
3986 * @addr: The DMA address
3987 * @size: The size of the region in bytes
3988 * @dir: The direction of the DMA
3990 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3993 enum dma_data_direction dir)
3995 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3999 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4000 * @dev: The device for which the DMA address was created
4001 * @addr: The DMA address
4002 * @size: The size of the region in bytes
4003 * @dir: The direction of the DMA
4005 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4008 enum dma_data_direction dir)
4010 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4014 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4015 * @dev: The device for which the DMA address is requested
4016 * @size: The size of the region to allocate in bytes
4017 * @dma_handle: A pointer for returning the DMA address of the region
4018 * @flag: memory allocator flags
4020 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4022 dma_addr_t *dma_handle,
4025 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4029 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4030 * @dev: The device for which the DMA addresses were allocated
4031 * @size: The size of the region
4032 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4033 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4035 static inline void ib_dma_free_coherent(struct ib_device *dev,
4036 size_t size, void *cpu_addr,
4037 dma_addr_t dma_handle)
4039 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4043 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4044 * HCA translation table.
4045 * @mr: The memory region to deregister.
4046 * @udata: Valid user data or NULL for kernel object
4048 * This function can fail, if the memory region has memory windows bound to it.
4050 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4053 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4054 * HCA translation table.
4055 * @mr: The memory region to deregister.
4057 * This function can fail, if the memory region has memory windows bound to it.
4059 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4061 static inline int ib_dereg_mr(struct ib_mr *mr)
4063 return ib_dereg_mr_user(mr, NULL);
4066 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4067 u32 max_num_sg, struct ib_udata *udata);
4069 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4070 enum ib_mr_type mr_type, u32 max_num_sg)
4072 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4075 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4076 u32 max_num_data_sg,
4077 u32 max_num_meta_sg);
4080 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4082 * @mr - struct ib_mr pointer to be updated.
4083 * @newkey - new key to be used.
4085 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4087 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4088 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4092 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4093 * for calculating a new rkey for type 2 memory windows.
4094 * @rkey - the rkey to increment.
4096 static inline u32 ib_inc_rkey(u32 rkey)
4098 const u32 mask = 0x000000ff;
4099 return ((rkey + 1) & mask) | (rkey & ~mask);
4103 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4104 * @pd: The protection domain associated with the unmapped region.
4105 * @mr_access_flags: Specifies the memory access rights.
4106 * @fmr_attr: Attributes of the unmapped region.
4108 * A fast memory region must be mapped before it can be used as part of
4111 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4112 int mr_access_flags,
4113 struct ib_fmr_attr *fmr_attr);
4116 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4117 * @fmr: The fast memory region to associate with the pages.
4118 * @page_list: An array of physical pages to map to the fast memory region.
4119 * @list_len: The number of pages in page_list.
4120 * @iova: The I/O virtual address to use with the mapped region.
4122 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4123 u64 *page_list, int list_len,
4126 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4130 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4131 * @fmr_list: A linked list of fast memory regions to unmap.
4133 int ib_unmap_fmr(struct list_head *fmr_list);
4136 * ib_dealloc_fmr - Deallocates a fast memory region.
4137 * @fmr: The fast memory region to deallocate.
4139 int ib_dealloc_fmr(struct ib_fmr *fmr);
4142 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4143 * @qp: QP to attach to the multicast group. The QP must be type
4145 * @gid: Multicast group GID.
4146 * @lid: Multicast group LID in host byte order.
4148 * In order to send and receive multicast packets, subnet
4149 * administration must have created the multicast group and configured
4150 * the fabric appropriately. The port associated with the specified
4151 * QP must also be a member of the multicast group.
4153 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4156 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4157 * @qp: QP to detach from the multicast group.
4158 * @gid: Multicast group GID.
4159 * @lid: Multicast group LID in host byte order.
4161 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4164 * ib_alloc_xrcd - Allocates an XRC domain.
4165 * @device: The device on which to allocate the XRC domain.
4166 * @caller: Module name for kernel consumers
4168 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4169 #define ib_alloc_xrcd(device) \
4170 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4173 * ib_dealloc_xrcd - Deallocates an XRC domain.
4174 * @xrcd: The XRC domain to deallocate.
4175 * @udata: Valid user data or NULL for kernel object
4177 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4179 static inline int ib_check_mr_access(int flags)
4182 * Local write permission is required if remote write or
4183 * remote atomic permission is also requested.
4185 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4186 !(flags & IB_ACCESS_LOCAL_WRITE))
4192 static inline bool ib_access_writable(int access_flags)
4195 * We have writable memory backing the MR if any of the following
4196 * access flags are set. "Local write" and "remote write" obviously
4197 * require write access. "Remote atomic" can do things like fetch and
4198 * add, which will modify memory, and "MW bind" can change permissions
4199 * by binding a window.
4201 return access_flags &
4202 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4203 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4207 * ib_check_mr_status: lightweight check of MR status.
4208 * This routine may provide status checks on a selected
4209 * ib_mr. first use is for signature status check.
4211 * @mr: A memory region.
4212 * @check_mask: Bitmask of which checks to perform from
4213 * ib_mr_status_check enumeration.
4214 * @mr_status: The container of relevant status checks.
4215 * failed checks will be indicated in the status bitmask
4216 * and the relevant info shall be in the error item.
4218 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4219 struct ib_mr_status *mr_status);
4222 * ib_device_try_get: Hold a registration lock
4223 * device: The device to lock
4225 * A device under an active registration lock cannot become unregistered. It
4226 * is only possible to obtain a registration lock on a device that is fully
4227 * registered, otherwise this function returns false.
4229 * The registration lock is only necessary for actions which require the
4230 * device to still be registered. Uses that only require the device pointer to
4231 * be valid should use get_device(&ibdev->dev) to hold the memory.
4234 static inline bool ib_device_try_get(struct ib_device *dev)
4236 return refcount_inc_not_zero(&dev->refcount);
4239 void ib_device_put(struct ib_device *device);
4240 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4241 enum rdma_driver_id driver_id);
4242 struct ib_device *ib_device_get_by_name(const char *name,
4243 enum rdma_driver_id driver_id);
4244 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4245 u16 pkey, const union ib_gid *gid,
4246 const struct sockaddr *addr);
4247 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4249 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4251 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4252 struct ib_wq_init_attr *init_attr);
4253 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4254 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4256 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4257 struct ib_rwq_ind_table_init_attr*
4258 wq_ind_table_init_attr);
4259 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4261 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4262 unsigned int *sg_offset, unsigned int page_size);
4263 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4264 int data_sg_nents, unsigned int *data_sg_offset,
4265 struct scatterlist *meta_sg, int meta_sg_nents,
4266 unsigned int *meta_sg_offset, unsigned int page_size);
4269 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4270 unsigned int *sg_offset, unsigned int page_size)
4274 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4280 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4281 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4283 void ib_drain_rq(struct ib_qp *qp);
4284 void ib_drain_sq(struct ib_qp *qp);
4285 void ib_drain_qp(struct ib_qp *qp);
4287 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4289 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4291 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4292 return attr->roce.dmac;
4296 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4298 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4299 attr->ib.dlid = (u16)dlid;
4300 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4301 attr->opa.dlid = dlid;
4304 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4306 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4307 return attr->ib.dlid;
4308 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4309 return attr->opa.dlid;
4313 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4318 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4323 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4326 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4327 attr->ib.src_path_bits = src_path_bits;
4328 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4329 attr->opa.src_path_bits = src_path_bits;
4332 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4334 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4335 return attr->ib.src_path_bits;
4336 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4337 return attr->opa.src_path_bits;
4341 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4344 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4345 attr->opa.make_grd = make_grd;
4348 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4350 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4351 return attr->opa.make_grd;
4355 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4357 attr->port_num = port_num;
4360 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4362 return attr->port_num;
4365 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4368 attr->static_rate = static_rate;
4371 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4373 return attr->static_rate;
4376 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4377 enum ib_ah_flags flag)
4379 attr->ah_flags = flag;
4382 static inline enum ib_ah_flags
4383 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4385 return attr->ah_flags;
4388 static inline const struct ib_global_route
4389 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4394 /*To retrieve and modify the grh */
4395 static inline struct ib_global_route
4396 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4401 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4403 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4405 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4408 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4411 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4413 grh->dgid.global.subnet_prefix = prefix;
4416 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4419 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4421 grh->dgid.global.interface_id = if_id;
4424 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4425 union ib_gid *dgid, u32 flow_label,
4426 u8 sgid_index, u8 hop_limit,
4429 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4431 attr->ah_flags = IB_AH_GRH;
4434 grh->flow_label = flow_label;
4435 grh->sgid_index = sgid_index;
4436 grh->hop_limit = hop_limit;
4437 grh->traffic_class = traffic_class;
4438 grh->sgid_attr = NULL;
4441 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4442 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4443 u32 flow_label, u8 hop_limit, u8 traffic_class,
4444 const struct ib_gid_attr *sgid_attr);
4445 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4446 const struct rdma_ah_attr *src);
4447 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4448 const struct rdma_ah_attr *new);
4449 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4452 * rdma_ah_find_type - Return address handle type.
4454 * @dev: Device to be checked
4455 * @port_num: Port number
4457 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4460 if (rdma_protocol_roce(dev, port_num))
4461 return RDMA_AH_ATTR_TYPE_ROCE;
4462 if (rdma_protocol_ib(dev, port_num)) {
4463 if (rdma_cap_opa_ah(dev, port_num))
4464 return RDMA_AH_ATTR_TYPE_OPA;
4465 return RDMA_AH_ATTR_TYPE_IB;
4468 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4472 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4473 * In the current implementation the only way to get
4474 * get the 32bit lid is from other sources for OPA.
4475 * For IB, lids will always be 16bits so cast the
4476 * value accordingly.
4480 static inline u16 ib_lid_cpu16(u32 lid)
4482 WARN_ON_ONCE(lid & 0xFFFF0000);
4487 * ib_lid_be16 - Return lid in 16bit BE encoding.
4491 static inline __be16 ib_lid_be16(u32 lid)
4493 WARN_ON_ONCE(lid & 0xFFFF0000);
4494 return cpu_to_be16((u16)lid);
4498 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4500 * @device: the rdma device
4501 * @comp_vector: index of completion vector
4503 * Returns NULL on failure, otherwise a corresponding cpu map of the
4504 * completion vector (returns all-cpus map if the device driver doesn't
4505 * implement get_vector_affinity).
4507 static inline const struct cpumask *
4508 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4510 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4511 !device->ops.get_vector_affinity)
4514 return device->ops.get_vector_affinity(device, comp_vector);
4519 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4520 * and add their gids, as needed, to the relevant RoCE devices.
4522 * @device: the rdma device
4524 void rdma_roce_rescan_device(struct ib_device *ibdev);
4526 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4528 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4530 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4531 enum rdma_netdev_t type, const char *name,
4532 unsigned char name_assign_type,
4533 void (*setup)(struct net_device *));
4535 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4536 enum rdma_netdev_t type, const char *name,
4537 unsigned char name_assign_type,
4538 void (*setup)(struct net_device *),
4539 struct net_device *netdev);
4542 * rdma_set_device_sysfs_group - Set device attributes group to have
4543 * driver specific sysfs entries at
4544 * for infiniband class.
4546 * @device: device pointer for which attributes to be created
4547 * @group: Pointer to group which should be added when device
4548 * is registered with sysfs.
4549 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4550 * group per device to have sysfs attributes.
4552 * NOTE: New drivers should not make use of this API; instead new device
4553 * parameter should be exposed via netlink command. This API and mechanism
4554 * exist only for existing drivers.
4557 rdma_set_device_sysfs_group(struct ib_device *dev,
4558 const struct attribute_group *group)
4560 dev->groups[1] = group;
4564 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4566 * @device: device pointer for which ib_device pointer to retrieve
4568 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4571 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4573 struct ib_core_device *coredev =
4574 container_of(device, struct ib_core_device, dev);
4576 return coredev->owner;
4580 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4581 * ib_device holder structure from device pointer.
4583 * NOTE: New drivers should not make use of this API; This API is only for
4584 * existing drivers who have exposed sysfs entries using
4585 * rdma_set_device_sysfs_group().
4587 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4588 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4590 bool rdma_dev_access_netns(const struct ib_device *device,
4591 const struct net *net);
4592 #endif /* IB_VERBS_H */