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 struct ib_usrq_object;
79 extern struct workqueue_struct *ib_wq;
80 extern struct workqueue_struct *ib_comp_wq;
81 extern struct workqueue_struct *ib_comp_unbound_wq;
86 void ibdev_printk(const char *level, const struct ib_device *ibdev,
87 const char *format, ...);
89 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
91 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
93 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
95 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
97 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
99 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
100 __printf(2, 3) __cold
101 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
103 #if defined(CONFIG_DYNAMIC_DEBUG)
104 #define ibdev_dbg(__dev, format, args...) \
105 dynamic_ibdev_dbg(__dev, format, ##args)
107 __printf(2, 3) __cold
109 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
112 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
114 static DEFINE_RATELIMIT_STATE(_rs, \
115 DEFAULT_RATELIMIT_INTERVAL, \
116 DEFAULT_RATELIMIT_BURST); \
117 if (__ratelimit(&_rs)) \
118 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
121 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
122 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
123 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
124 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
125 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
126 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
127 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
128 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
129 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
130 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
131 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
132 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
133 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
134 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
136 #if defined(CONFIG_DYNAMIC_DEBUG)
137 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
138 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
140 static DEFINE_RATELIMIT_STATE(_rs, \
141 DEFAULT_RATELIMIT_INTERVAL, \
142 DEFAULT_RATELIMIT_BURST); \
143 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
144 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
145 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
149 __printf(2, 3) __cold
151 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
157 __be64 subnet_prefix;
162 extern union ib_gid zgid;
165 /* If link layer is Ethernet, this is RoCE V1 */
167 IB_GID_TYPE_ROCE = 0,
168 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
172 #define ROCE_V2_UDP_DPORT 4791
174 struct net_device __rcu *ndev;
175 struct ib_device *device;
177 enum ib_gid_type gid_type;
183 /* set the local administered indication */
184 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
187 enum rdma_transport_type {
189 RDMA_TRANSPORT_IWARP,
190 RDMA_TRANSPORT_USNIC,
191 RDMA_TRANSPORT_USNIC_UDP,
192 RDMA_TRANSPORT_UNSPECIFIED,
195 enum rdma_protocol_type {
199 RDMA_PROTOCOL_USNIC_UDP
202 __attribute_const__ enum rdma_transport_type
203 rdma_node_get_transport(unsigned int node_type);
205 enum rdma_network_type {
207 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
212 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
214 if (network_type == RDMA_NETWORK_IPV4 ||
215 network_type == RDMA_NETWORK_IPV6)
216 return IB_GID_TYPE_ROCE_UDP_ENCAP;
218 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
219 return IB_GID_TYPE_IB;
222 static inline enum rdma_network_type
223 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
225 if (attr->gid_type == IB_GID_TYPE_IB)
226 return RDMA_NETWORK_IB;
228 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
229 return RDMA_NETWORK_IPV4;
231 return RDMA_NETWORK_IPV6;
234 enum rdma_link_layer {
235 IB_LINK_LAYER_UNSPECIFIED,
236 IB_LINK_LAYER_INFINIBAND,
237 IB_LINK_LAYER_ETHERNET,
240 enum ib_device_cap_flags {
241 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
242 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
243 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
244 IB_DEVICE_RAW_MULTI = (1 << 3),
245 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
246 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
247 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
248 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
249 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
250 /* Not in use, former INIT_TYPE = (1 << 9),*/
251 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
252 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
253 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
254 IB_DEVICE_SRQ_RESIZE = (1 << 13),
255 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
258 * This device supports a per-device lkey or stag that can be
259 * used without performing a memory registration for the local
260 * memory. Note that ULPs should never check this flag, but
261 * instead of use the local_dma_lkey flag in the ib_pd structure,
262 * which will always contain a usable lkey.
264 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
265 /* Reserved, old SEND_W_INV = (1 << 16),*/
266 IB_DEVICE_MEM_WINDOW = (1 << 17),
268 * Devices should set IB_DEVICE_UD_IP_SUM if they support
269 * insertion of UDP and TCP checksum on outgoing UD IPoIB
270 * messages and can verify the validity of checksum for
271 * incoming messages. Setting this flag implies that the
272 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
274 IB_DEVICE_UD_IP_CSUM = (1 << 18),
275 IB_DEVICE_UD_TSO = (1 << 19),
276 IB_DEVICE_XRC = (1 << 20),
279 * This device supports the IB "base memory management extension",
280 * which includes support for fast registrations (IB_WR_REG_MR,
281 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
282 * also be set by any iWarp device which must support FRs to comply
283 * to the iWarp verbs spec. iWarp devices also support the
284 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
287 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
288 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
289 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
290 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
291 IB_DEVICE_RC_IP_CSUM = (1 << 25),
292 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
293 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
295 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
296 * support execution of WQEs that involve synchronization
297 * of I/O operations with single completion queue managed
300 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
301 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
302 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
303 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
304 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
305 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
306 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
307 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
308 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
309 /* The device supports padding incoming writes to cacheline. */
310 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
311 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
320 enum ib_odp_general_cap_bits {
321 IB_ODP_SUPPORT = 1 << 0,
322 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
325 enum ib_odp_transport_cap_bits {
326 IB_ODP_SUPPORT_SEND = 1 << 0,
327 IB_ODP_SUPPORT_RECV = 1 << 1,
328 IB_ODP_SUPPORT_WRITE = 1 << 2,
329 IB_ODP_SUPPORT_READ = 1 << 3,
330 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
331 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
335 uint64_t general_caps;
337 uint32_t rc_odp_caps;
338 uint32_t uc_odp_caps;
339 uint32_t ud_odp_caps;
340 uint32_t xrc_odp_caps;
341 } per_transport_caps;
345 /* Corresponding bit will be set if qp type from
346 * 'enum ib_qp_type' is supported, e.g.
347 * supported_qpts |= 1 << IB_QPT_UD
350 u32 max_rwq_indirection_tables;
351 u32 max_rwq_indirection_table_size;
354 enum ib_tm_cap_flags {
355 /* Support tag matching with rendezvous offload for RC transport */
356 IB_TM_CAP_RNDV_RC = 1 << 0,
360 /* Max size of RNDV header */
361 u32 max_rndv_hdr_size;
362 /* Max number of entries in tag matching list */
364 /* From enum ib_tm_cap_flags */
366 /* Max number of outstanding list operations */
368 /* Max number of SGE in tag matching entry */
372 struct ib_cq_init_attr {
378 enum ib_cq_attr_mask {
379 IB_CQ_MODERATE = 1 << 0,
383 u16 max_cq_moderation_count;
384 u16 max_cq_moderation_period;
387 struct ib_dm_mr_attr {
393 struct ib_dm_alloc_attr {
399 struct ib_device_attr {
401 __be64 sys_image_guid;
409 u64 device_cap_flags;
420 int max_qp_init_rd_atom;
421 int max_ee_init_rd_atom;
422 enum ib_atomic_cap atomic_cap;
423 enum ib_atomic_cap masked_atomic_cap;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
438 unsigned int max_fast_reg_page_list_len;
439 unsigned int max_pi_fast_reg_page_list_len;
441 u8 local_ca_ack_delay;
444 struct ib_odp_caps odp_caps;
445 uint64_t timestamp_mask;
446 uint64_t hca_core_clock; /* in KHZ */
447 struct ib_rss_caps rss_caps;
449 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
450 struct ib_tm_caps tm_caps;
451 struct ib_cq_caps cq_caps;
453 /* Max entries for sgl for optimized performance per READ */
465 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
468 case IB_MTU_256: return 256;
469 case IB_MTU_512: return 512;
470 case IB_MTU_1024: return 1024;
471 case IB_MTU_2048: return 2048;
472 case IB_MTU_4096: return 4096;
477 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
481 else if (mtu >= 2048)
483 else if (mtu >= 1024)
497 IB_PORT_ACTIVE_DEFER = 5
500 enum ib_port_phys_state {
501 IB_PORT_PHYS_STATE_SLEEP = 1,
502 IB_PORT_PHYS_STATE_POLLING = 2,
503 IB_PORT_PHYS_STATE_DISABLED = 3,
504 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
505 IB_PORT_PHYS_STATE_LINK_UP = 5,
506 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
507 IB_PORT_PHYS_STATE_PHY_TEST = 7,
518 static inline int ib_width_enum_to_int(enum ib_port_width width)
521 case IB_WIDTH_1X: return 1;
522 case IB_WIDTH_2X: return 2;
523 case IB_WIDTH_4X: return 4;
524 case IB_WIDTH_8X: return 8;
525 case IB_WIDTH_12X: return 12;
541 * struct rdma_hw_stats
542 * @lock - Mutex to protect parallel write access to lifespan and values
543 * of counters, which are 64bits and not guaranteeed to be written
544 * atomicaly on 32bits systems.
545 * @timestamp - Used by the core code to track when the last update was
546 * @lifespan - Used by the core code to determine how old the counters
547 * should be before being updated again. Stored in jiffies, defaults
548 * to 10 milliseconds, drivers can override the default be specifying
549 * their own value during their allocation routine.
550 * @name - Array of pointers to static names used for the counters in
552 * @num_counters - How many hardware counters there are. If name is
553 * shorter than this number, a kernel oops will result. Driver authors
554 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
555 * in their code to prevent this.
556 * @value - Array of u64 counters that are accessed by the sysfs code and
557 * filled in by the drivers get_stats routine
559 struct rdma_hw_stats {
560 struct mutex lock; /* Protect lifespan and values[] */
561 unsigned long timestamp;
562 unsigned long lifespan;
563 const char * const *names;
568 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
570 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
572 * @names - Array of static const char *
573 * @num_counters - How many elements in array
574 * @lifespan - How many milliseconds between updates
576 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
577 const char * const *names, int num_counters,
578 unsigned long lifespan)
580 struct rdma_hw_stats *stats;
582 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
586 stats->names = names;
587 stats->num_counters = num_counters;
588 stats->lifespan = msecs_to_jiffies(lifespan);
594 /* Define bits for the various functionality this port needs to be supported by
597 /* Management 0x00000FFF */
598 #define RDMA_CORE_CAP_IB_MAD 0x00000001
599 #define RDMA_CORE_CAP_IB_SMI 0x00000002
600 #define RDMA_CORE_CAP_IB_CM 0x00000004
601 #define RDMA_CORE_CAP_IW_CM 0x00000008
602 #define RDMA_CORE_CAP_IB_SA 0x00000010
603 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
605 /* Address format 0x000FF000 */
606 #define RDMA_CORE_CAP_AF_IB 0x00001000
607 #define RDMA_CORE_CAP_ETH_AH 0x00002000
608 #define RDMA_CORE_CAP_OPA_AH 0x00004000
609 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
611 /* Protocol 0xFFF00000 */
612 #define RDMA_CORE_CAP_PROT_IB 0x00100000
613 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
614 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
615 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
616 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
617 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
619 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
620 | RDMA_CORE_CAP_PROT_ROCE \
621 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
623 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
624 | RDMA_CORE_CAP_IB_MAD \
625 | RDMA_CORE_CAP_IB_SMI \
626 | RDMA_CORE_CAP_IB_CM \
627 | RDMA_CORE_CAP_IB_SA \
628 | RDMA_CORE_CAP_AF_IB)
629 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
630 | RDMA_CORE_CAP_IB_MAD \
631 | RDMA_CORE_CAP_IB_CM \
632 | RDMA_CORE_CAP_AF_IB \
633 | RDMA_CORE_CAP_ETH_AH)
634 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
635 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
636 | RDMA_CORE_CAP_IB_MAD \
637 | RDMA_CORE_CAP_IB_CM \
638 | RDMA_CORE_CAP_AF_IB \
639 | RDMA_CORE_CAP_ETH_AH)
640 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
641 | RDMA_CORE_CAP_IW_CM)
642 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
643 | RDMA_CORE_CAP_OPA_MAD)
645 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
647 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
649 struct ib_port_attr {
651 enum ib_port_state state;
653 enum ib_mtu active_mtu;
655 unsigned int ip_gids:1;
656 /* This is the value from PortInfo CapabilityMask, defined by IBA */
675 enum ib_device_modify_flags {
676 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
677 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
680 #define IB_DEVICE_NODE_DESC_MAX 64
682 struct ib_device_modify {
684 char node_desc[IB_DEVICE_NODE_DESC_MAX];
687 enum ib_port_modify_flags {
688 IB_PORT_SHUTDOWN = 1,
689 IB_PORT_INIT_TYPE = (1<<2),
690 IB_PORT_RESET_QKEY_CNTR = (1<<3),
691 IB_PORT_OPA_MASK_CHG = (1<<4)
694 struct ib_port_modify {
695 u32 set_port_cap_mask;
696 u32 clr_port_cap_mask;
704 IB_EVENT_QP_ACCESS_ERR,
708 IB_EVENT_PATH_MIG_ERR,
709 IB_EVENT_DEVICE_FATAL,
710 IB_EVENT_PORT_ACTIVE,
713 IB_EVENT_PKEY_CHANGE,
716 IB_EVENT_SRQ_LIMIT_REACHED,
717 IB_EVENT_QP_LAST_WQE_REACHED,
718 IB_EVENT_CLIENT_REREGISTER,
723 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
726 struct ib_device *device;
734 enum ib_event_type event;
737 struct ib_event_handler {
738 struct ib_device *device;
739 void (*handler)(struct ib_event_handler *, struct ib_event *);
740 struct list_head list;
743 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
745 (_ptr)->device = _device; \
746 (_ptr)->handler = _handler; \
747 INIT_LIST_HEAD(&(_ptr)->list); \
750 struct ib_global_route {
751 const struct ib_gid_attr *sgid_attr;
760 __be32 version_tclass_flow;
768 union rdma_network_hdr {
771 /* The IB spec states that if it's IPv4, the header
772 * is located in the last 20 bytes of the header.
775 struct iphdr roce4grh;
779 #define IB_QPN_MASK 0xFFFFFF
782 IB_MULTICAST_QPN = 0xffffff
785 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
786 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
793 IB_RATE_PORT_CURRENT = 0,
794 IB_RATE_2_5_GBPS = 2,
802 IB_RATE_120_GBPS = 10,
803 IB_RATE_14_GBPS = 11,
804 IB_RATE_56_GBPS = 12,
805 IB_RATE_112_GBPS = 13,
806 IB_RATE_168_GBPS = 14,
807 IB_RATE_25_GBPS = 15,
808 IB_RATE_100_GBPS = 16,
809 IB_RATE_200_GBPS = 17,
810 IB_RATE_300_GBPS = 18,
811 IB_RATE_28_GBPS = 19,
812 IB_RATE_50_GBPS = 20,
813 IB_RATE_400_GBPS = 21,
814 IB_RATE_600_GBPS = 22,
818 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
819 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
820 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
821 * @rate: rate to convert.
823 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
826 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
827 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
828 * @rate: rate to convert.
830 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
834 * enum ib_mr_type - memory region type
835 * @IB_MR_TYPE_MEM_REG: memory region that is used for
836 * normal registration
837 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
838 * register any arbitrary sg lists (without
839 * the normal mr constraints - see
841 * @IB_MR_TYPE_DM: memory region that is used for device
842 * memory registration
843 * @IB_MR_TYPE_USER: memory region that is used for the user-space
845 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
846 * without address translations (VA=PA)
847 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
848 * data integrity operations
856 IB_MR_TYPE_INTEGRITY,
859 enum ib_mr_status_check {
860 IB_MR_CHECK_SIG_STATUS = 1,
864 * struct ib_mr_status - Memory region status container
866 * @fail_status: Bitmask of MR checks status. For each
867 * failed check a corresponding status bit is set.
868 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
871 struct ib_mr_status {
873 struct ib_sig_err sig_err;
877 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
879 * @mult: multiple to convert.
881 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
883 enum rdma_ah_attr_type {
884 RDMA_AH_ATTR_TYPE_UNDEFINED,
885 RDMA_AH_ATTR_TYPE_IB,
886 RDMA_AH_ATTR_TYPE_ROCE,
887 RDMA_AH_ATTR_TYPE_OPA,
895 struct roce_ah_attr {
905 struct rdma_ah_attr {
906 struct ib_global_route grh;
911 enum rdma_ah_attr_type type;
913 struct ib_ah_attr ib;
914 struct roce_ah_attr roce;
915 struct opa_ah_attr opa;
923 IB_WC_LOC_EEC_OP_ERR,
928 IB_WC_LOC_ACCESS_ERR,
929 IB_WC_REM_INV_REQ_ERR,
930 IB_WC_REM_ACCESS_ERR,
933 IB_WC_RNR_RETRY_EXC_ERR,
934 IB_WC_LOC_RDD_VIOL_ERR,
935 IB_WC_REM_INV_RD_REQ_ERR,
938 IB_WC_INV_EEC_STATE_ERR,
940 IB_WC_RESP_TIMEOUT_ERR,
944 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
955 IB_WC_MASKED_COMP_SWAP,
956 IB_WC_MASKED_FETCH_ADD,
958 * Set value of IB_WC_RECV so consumers can test if a completion is a
959 * receive by testing (opcode & IB_WC_RECV).
962 IB_WC_RECV_RDMA_WITH_IMM
967 IB_WC_WITH_IMM = (1<<1),
968 IB_WC_WITH_INVALIDATE = (1<<2),
969 IB_WC_IP_CSUM_OK = (1<<3),
970 IB_WC_WITH_SMAC = (1<<4),
971 IB_WC_WITH_VLAN = (1<<5),
972 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
978 struct ib_cqe *wr_cqe;
980 enum ib_wc_status status;
981 enum ib_wc_opcode opcode;
995 u8 port_num; /* valid only for DR SMPs on switches */
1001 enum ib_cq_notify_flags {
1002 IB_CQ_SOLICITED = 1 << 0,
1003 IB_CQ_NEXT_COMP = 1 << 1,
1004 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1005 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1014 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1016 return srq_type == IB_SRQT_XRC ||
1017 srq_type == IB_SRQT_TM;
1020 enum ib_srq_attr_mask {
1021 IB_SRQ_MAX_WR = 1 << 0,
1022 IB_SRQ_LIMIT = 1 << 1,
1025 struct ib_srq_attr {
1031 struct ib_srq_init_attr {
1032 void (*event_handler)(struct ib_event *, void *);
1034 struct ib_srq_attr attr;
1035 enum ib_srq_type srq_type;
1041 struct ib_xrcd *xrcd;
1056 u32 max_inline_data;
1059 * Maximum number of rdma_rw_ctx structures in flight at a time.
1060 * ib_create_qp() will calculate the right amount of neededed WRs
1061 * and MRs based on this.
1073 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1074 * here (and in that order) since the MAD layer uses them as
1075 * indices into a 2-entry table.
1084 IB_QPT_RAW_ETHERTYPE,
1085 IB_QPT_RAW_PACKET = 8,
1089 IB_QPT_DRIVER = 0xFF,
1090 /* Reserve a range for qp types internal to the low level driver.
1091 * These qp types will not be visible at the IB core layer, so the
1092 * IB_QPT_MAX usages should not be affected in the core layer
1094 IB_QPT_RESERVED1 = 0x1000,
1106 enum ib_qp_create_flags {
1107 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1108 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1109 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1110 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1111 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1112 IB_QP_CREATE_NETIF_QP = 1 << 5,
1113 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1114 /* FREE = 1 << 7, */
1115 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1116 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1117 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1118 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1119 /* reserve bits 26-31 for low level drivers' internal use */
1120 IB_QP_CREATE_RESERVED_START = 1 << 26,
1121 IB_QP_CREATE_RESERVED_END = 1 << 31,
1125 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1126 * callback to destroy the passed in QP.
1129 struct ib_qp_init_attr {
1130 /* Consumer's event_handler callback must not block */
1131 void (*event_handler)(struct ib_event *, void *);
1134 struct ib_cq *send_cq;
1135 struct ib_cq *recv_cq;
1137 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1138 struct ib_qp_cap cap;
1139 enum ib_sig_type sq_sig_type;
1140 enum ib_qp_type qp_type;
1144 * Only needed for special QP types, or when using the RW API.
1147 struct ib_rwq_ind_table *rwq_ind_tbl;
1151 struct ib_qp_open_attr {
1152 void (*event_handler)(struct ib_event *, void *);
1155 enum ib_qp_type qp_type;
1158 enum ib_rnr_timeout {
1159 IB_RNR_TIMER_655_36 = 0,
1160 IB_RNR_TIMER_000_01 = 1,
1161 IB_RNR_TIMER_000_02 = 2,
1162 IB_RNR_TIMER_000_03 = 3,
1163 IB_RNR_TIMER_000_04 = 4,
1164 IB_RNR_TIMER_000_06 = 5,
1165 IB_RNR_TIMER_000_08 = 6,
1166 IB_RNR_TIMER_000_12 = 7,
1167 IB_RNR_TIMER_000_16 = 8,
1168 IB_RNR_TIMER_000_24 = 9,
1169 IB_RNR_TIMER_000_32 = 10,
1170 IB_RNR_TIMER_000_48 = 11,
1171 IB_RNR_TIMER_000_64 = 12,
1172 IB_RNR_TIMER_000_96 = 13,
1173 IB_RNR_TIMER_001_28 = 14,
1174 IB_RNR_TIMER_001_92 = 15,
1175 IB_RNR_TIMER_002_56 = 16,
1176 IB_RNR_TIMER_003_84 = 17,
1177 IB_RNR_TIMER_005_12 = 18,
1178 IB_RNR_TIMER_007_68 = 19,
1179 IB_RNR_TIMER_010_24 = 20,
1180 IB_RNR_TIMER_015_36 = 21,
1181 IB_RNR_TIMER_020_48 = 22,
1182 IB_RNR_TIMER_030_72 = 23,
1183 IB_RNR_TIMER_040_96 = 24,
1184 IB_RNR_TIMER_061_44 = 25,
1185 IB_RNR_TIMER_081_92 = 26,
1186 IB_RNR_TIMER_122_88 = 27,
1187 IB_RNR_TIMER_163_84 = 28,
1188 IB_RNR_TIMER_245_76 = 29,
1189 IB_RNR_TIMER_327_68 = 30,
1190 IB_RNR_TIMER_491_52 = 31
1193 enum ib_qp_attr_mask {
1195 IB_QP_CUR_STATE = (1<<1),
1196 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1197 IB_QP_ACCESS_FLAGS = (1<<3),
1198 IB_QP_PKEY_INDEX = (1<<4),
1199 IB_QP_PORT = (1<<5),
1200 IB_QP_QKEY = (1<<6),
1202 IB_QP_PATH_MTU = (1<<8),
1203 IB_QP_TIMEOUT = (1<<9),
1204 IB_QP_RETRY_CNT = (1<<10),
1205 IB_QP_RNR_RETRY = (1<<11),
1206 IB_QP_RQ_PSN = (1<<12),
1207 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1208 IB_QP_ALT_PATH = (1<<14),
1209 IB_QP_MIN_RNR_TIMER = (1<<15),
1210 IB_QP_SQ_PSN = (1<<16),
1211 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1212 IB_QP_PATH_MIG_STATE = (1<<18),
1213 IB_QP_CAP = (1<<19),
1214 IB_QP_DEST_QPN = (1<<20),
1215 IB_QP_RESERVED1 = (1<<21),
1216 IB_QP_RESERVED2 = (1<<22),
1217 IB_QP_RESERVED3 = (1<<23),
1218 IB_QP_RESERVED4 = (1<<24),
1219 IB_QP_RATE_LIMIT = (1<<25),
1244 enum ib_qp_state qp_state;
1245 enum ib_qp_state cur_qp_state;
1246 enum ib_mtu path_mtu;
1247 enum ib_mig_state path_mig_state;
1252 int qp_access_flags;
1253 struct ib_qp_cap cap;
1254 struct rdma_ah_attr ah_attr;
1255 struct rdma_ah_attr alt_ah_attr;
1258 u8 en_sqd_async_notify;
1261 u8 max_dest_rd_atomic;
1273 /* These are shared with userspace */
1274 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1275 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1276 IB_WR_SEND = IB_UVERBS_WR_SEND,
1277 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1278 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1279 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1280 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1281 IB_WR_LSO = IB_UVERBS_WR_TSO,
1282 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1283 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1284 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1285 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1286 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1287 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1288 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1290 /* These are kernel only and can not be issued by userspace */
1291 IB_WR_REG_MR = 0x20,
1292 IB_WR_REG_MR_INTEGRITY,
1294 /* reserve values for low level drivers' internal use.
1295 * These values will not be used at all in the ib core layer.
1297 IB_WR_RESERVED1 = 0xf0,
1309 enum ib_send_flags {
1311 IB_SEND_SIGNALED = (1<<1),
1312 IB_SEND_SOLICITED = (1<<2),
1313 IB_SEND_INLINE = (1<<3),
1314 IB_SEND_IP_CSUM = (1<<4),
1316 /* reserve bits 26-31 for low level drivers' internal use */
1317 IB_SEND_RESERVED_START = (1 << 26),
1318 IB_SEND_RESERVED_END = (1 << 31),
1328 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1332 struct ib_send_wr *next;
1335 struct ib_cqe *wr_cqe;
1337 struct ib_sge *sg_list;
1339 enum ib_wr_opcode opcode;
1343 u32 invalidate_rkey;
1348 struct ib_send_wr wr;
1353 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1355 return container_of(wr, struct ib_rdma_wr, wr);
1358 struct ib_atomic_wr {
1359 struct ib_send_wr wr;
1363 u64 compare_add_mask;
1368 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1370 return container_of(wr, struct ib_atomic_wr, wr);
1374 struct ib_send_wr wr;
1381 u16 pkey_index; /* valid for GSI only */
1382 u8 port_num; /* valid for DR SMPs on switch only */
1385 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1387 return container_of(wr, struct ib_ud_wr, wr);
1391 struct ib_send_wr wr;
1397 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1399 return container_of(wr, struct ib_reg_wr, wr);
1403 struct ib_recv_wr *next;
1406 struct ib_cqe *wr_cqe;
1408 struct ib_sge *sg_list;
1412 enum ib_access_flags {
1413 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1414 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1415 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1416 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1417 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1418 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1419 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1420 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1421 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1423 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1424 IB_ACCESS_SUPPORTED =
1425 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1429 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1430 * are hidden here instead of a uapi header!
1432 enum ib_mr_rereg_flags {
1433 IB_MR_REREG_TRANS = 1,
1434 IB_MR_REREG_PD = (1<<1),
1435 IB_MR_REREG_ACCESS = (1<<2),
1436 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1439 struct ib_fmr_attr {
1447 enum rdma_remove_reason {
1449 * Userspace requested uobject deletion or initial try
1450 * to remove uobject via cleanup. Call could fail
1452 RDMA_REMOVE_DESTROY,
1453 /* Context deletion. This call should delete the actual object itself */
1455 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1456 RDMA_REMOVE_DRIVER_REMOVE,
1457 /* uobj is being cleaned-up before being committed */
1461 struct ib_rdmacg_object {
1462 #ifdef CONFIG_CGROUP_RDMA
1463 struct rdma_cgroup *cg; /* owner rdma cgroup */
1467 struct ib_ucontext {
1468 struct ib_device *device;
1469 struct ib_uverbs_file *ufile;
1471 * 'closing' can be read by the driver only during a destroy callback,
1472 * it is set when we are closing the file descriptor and indicates
1473 * that mm_sem may be locked.
1477 bool cleanup_retryable;
1479 struct ib_rdmacg_object cg_obj;
1481 * Implementation details of the RDMA core, don't use in drivers:
1483 struct rdma_restrack_entry res;
1484 struct xarray mmap_xa;
1488 u64 user_handle; /* handle given to us by userspace */
1489 /* ufile & ucontext owning this object */
1490 struct ib_uverbs_file *ufile;
1491 /* FIXME, save memory: ufile->context == context */
1492 struct ib_ucontext *context; /* associated user context */
1493 void *object; /* containing object */
1494 struct list_head list; /* link to context's list */
1495 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1496 int id; /* index into kernel idr */
1498 atomic_t usecnt; /* protects exclusive access */
1499 struct rcu_head rcu; /* kfree_rcu() overhead */
1501 const struct uverbs_api_object *uapi_object;
1505 const void __user *inbuf;
1506 void __user *outbuf;
1514 struct ib_device *device;
1515 struct ib_uobject *uobject;
1516 atomic_t usecnt; /* count all resources */
1518 u32 unsafe_global_rkey;
1521 * Implementation details of the RDMA core, don't use in drivers:
1523 struct ib_mr *__internal_mr;
1524 struct rdma_restrack_entry res;
1528 struct ib_device *device;
1529 atomic_t usecnt; /* count all exposed resources */
1530 struct inode *inode;
1532 struct mutex tgt_qp_mutex;
1533 struct list_head tgt_qp_list;
1537 struct ib_device *device;
1539 struct ib_uobject *uobject;
1540 const struct ib_gid_attr *sgid_attr;
1541 enum rdma_ah_attr_type type;
1544 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1546 enum ib_poll_context {
1547 IB_POLL_DIRECT, /* caller context, no hw completions */
1548 IB_POLL_SOFTIRQ, /* poll from softirq context */
1549 IB_POLL_WORKQUEUE, /* poll from workqueue */
1550 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1554 struct ib_device *device;
1555 struct ib_ucq_object *uobject;
1556 ib_comp_handler comp_handler;
1557 void (*event_handler)(struct ib_event *, void *);
1560 atomic_t usecnt; /* count number of work queues */
1561 enum ib_poll_context poll_ctx;
1564 struct irq_poll iop;
1565 struct work_struct work;
1567 struct workqueue_struct *comp_wq;
1570 /* updated only by trace points */
1575 * Implementation details of the RDMA core, don't use in drivers:
1577 struct rdma_restrack_entry res;
1581 struct ib_device *device;
1583 struct ib_usrq_object *uobject;
1584 void (*event_handler)(struct ib_event *, void *);
1586 enum ib_srq_type srq_type;
1593 struct ib_xrcd *xrcd;
1600 enum ib_raw_packet_caps {
1601 /* Strip cvlan from incoming packet and report it in the matching work
1602 * completion is supported.
1604 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1605 /* Scatter FCS field of an incoming packet to host memory is supported.
1607 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1608 /* Checksum offloads are supported (for both send and receive). */
1609 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1610 /* When a packet is received for an RQ with no receive WQEs, the
1611 * packet processing is delayed.
1613 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1627 struct ib_device *device;
1628 struct ib_uwq_object *uobject;
1630 void (*event_handler)(struct ib_event *, void *);
1634 enum ib_wq_state state;
1635 enum ib_wq_type wq_type;
1640 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1641 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1642 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1643 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1646 struct ib_wq_init_attr {
1648 enum ib_wq_type wq_type;
1652 void (*event_handler)(struct ib_event *, void *);
1653 u32 create_flags; /* Use enum ib_wq_flags */
1656 enum ib_wq_attr_mask {
1657 IB_WQ_STATE = 1 << 0,
1658 IB_WQ_CUR_STATE = 1 << 1,
1659 IB_WQ_FLAGS = 1 << 2,
1663 enum ib_wq_state wq_state;
1664 enum ib_wq_state curr_wq_state;
1665 u32 flags; /* Use enum ib_wq_flags */
1666 u32 flags_mask; /* Use enum ib_wq_flags */
1669 struct ib_rwq_ind_table {
1670 struct ib_device *device;
1671 struct ib_uobject *uobject;
1674 u32 log_ind_tbl_size;
1675 struct ib_wq **ind_tbl;
1678 struct ib_rwq_ind_table_init_attr {
1679 u32 log_ind_tbl_size;
1680 /* Each entry is a pointer to Receive Work Queue */
1681 struct ib_wq **ind_tbl;
1684 enum port_pkey_state {
1685 IB_PORT_PKEY_NOT_VALID = 0,
1686 IB_PORT_PKEY_VALID = 1,
1687 IB_PORT_PKEY_LISTED = 2,
1690 struct ib_qp_security;
1692 struct ib_port_pkey {
1693 enum port_pkey_state state;
1696 struct list_head qp_list;
1697 struct list_head to_error_list;
1698 struct ib_qp_security *sec;
1701 struct ib_ports_pkeys {
1702 struct ib_port_pkey main;
1703 struct ib_port_pkey alt;
1706 struct ib_qp_security {
1708 struct ib_device *dev;
1709 /* Hold this mutex when changing port and pkey settings. */
1711 struct ib_ports_pkeys *ports_pkeys;
1712 /* A list of all open shared QP handles. Required to enforce security
1713 * properly for all users of a shared QP.
1715 struct list_head shared_qp_list;
1718 atomic_t error_list_count;
1719 struct completion error_complete;
1720 int error_comps_pending;
1724 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1725 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1728 struct ib_device *device;
1730 struct ib_cq *send_cq;
1731 struct ib_cq *recv_cq;
1734 struct list_head rdma_mrs;
1735 struct list_head sig_mrs;
1737 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1738 struct list_head xrcd_list;
1740 /* count times opened, mcast attaches, flow attaches */
1742 struct list_head open_list;
1743 struct ib_qp *real_qp;
1744 struct ib_uqp_object *uobject;
1745 void (*event_handler)(struct ib_event *, void *);
1747 /* sgid_attrs associated with the AV's */
1748 const struct ib_gid_attr *av_sgid_attr;
1749 const struct ib_gid_attr *alt_path_sgid_attr;
1753 enum ib_qp_type qp_type;
1754 struct ib_rwq_ind_table *rwq_ind_tbl;
1755 struct ib_qp_security *qp_sec;
1760 * Implementation details of the RDMA core, don't use in drivers:
1762 struct rdma_restrack_entry res;
1764 /* The counter the qp is bind to */
1765 struct rdma_counter *counter;
1769 struct ib_device *device;
1772 struct ib_uobject *uobject;
1777 struct ib_device *device;
1783 unsigned int page_size;
1784 enum ib_mr_type type;
1787 struct ib_uobject *uobject; /* user */
1788 struct list_head qp_entry; /* FR */
1792 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1794 * Implementation details of the RDMA core, don't use in drivers:
1796 struct rdma_restrack_entry res;
1800 struct ib_device *device;
1802 struct ib_uobject *uobject;
1804 enum ib_mw_type type;
1808 struct ib_device *device;
1810 struct list_head list;
1815 /* Supported steering options */
1816 enum ib_flow_attr_type {
1817 /* steering according to rule specifications */
1818 IB_FLOW_ATTR_NORMAL = 0x0,
1819 /* default unicast and multicast rule -
1820 * receive all Eth traffic which isn't steered to any QP
1822 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1823 /* default multicast rule -
1824 * receive all Eth multicast traffic which isn't steered to any QP
1826 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1827 /* sniffer rule - receive all port traffic */
1828 IB_FLOW_ATTR_SNIFFER = 0x3
1831 /* Supported steering header types */
1832 enum ib_flow_spec_type {
1834 IB_FLOW_SPEC_ETH = 0x20,
1835 IB_FLOW_SPEC_IB = 0x22,
1837 IB_FLOW_SPEC_IPV4 = 0x30,
1838 IB_FLOW_SPEC_IPV6 = 0x31,
1839 IB_FLOW_SPEC_ESP = 0x34,
1841 IB_FLOW_SPEC_TCP = 0x40,
1842 IB_FLOW_SPEC_UDP = 0x41,
1843 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1844 IB_FLOW_SPEC_GRE = 0x51,
1845 IB_FLOW_SPEC_MPLS = 0x60,
1846 IB_FLOW_SPEC_INNER = 0x100,
1848 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1849 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1850 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1851 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1853 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1854 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1856 /* Flow steering rule priority is set according to it's domain.
1857 * Lower domain value means higher priority.
1859 enum ib_flow_domain {
1860 IB_FLOW_DOMAIN_USER,
1861 IB_FLOW_DOMAIN_ETHTOOL,
1864 IB_FLOW_DOMAIN_NUM /* Must be last */
1867 enum ib_flow_flags {
1868 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1869 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1870 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1873 struct ib_flow_eth_filter {
1882 struct ib_flow_spec_eth {
1885 struct ib_flow_eth_filter val;
1886 struct ib_flow_eth_filter mask;
1889 struct ib_flow_ib_filter {
1896 struct ib_flow_spec_ib {
1899 struct ib_flow_ib_filter val;
1900 struct ib_flow_ib_filter mask;
1903 /* IPv4 header flags */
1904 enum ib_ipv4_flags {
1905 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1906 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1907 last have this flag set */
1910 struct ib_flow_ipv4_filter {
1921 struct ib_flow_spec_ipv4 {
1924 struct ib_flow_ipv4_filter val;
1925 struct ib_flow_ipv4_filter mask;
1928 struct ib_flow_ipv6_filter {
1939 struct ib_flow_spec_ipv6 {
1942 struct ib_flow_ipv6_filter val;
1943 struct ib_flow_ipv6_filter mask;
1946 struct ib_flow_tcp_udp_filter {
1953 struct ib_flow_spec_tcp_udp {
1956 struct ib_flow_tcp_udp_filter val;
1957 struct ib_flow_tcp_udp_filter mask;
1960 struct ib_flow_tunnel_filter {
1965 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1966 * the tunnel_id from val has the vni value
1968 struct ib_flow_spec_tunnel {
1971 struct ib_flow_tunnel_filter val;
1972 struct ib_flow_tunnel_filter mask;
1975 struct ib_flow_esp_filter {
1982 struct ib_flow_spec_esp {
1985 struct ib_flow_esp_filter val;
1986 struct ib_flow_esp_filter mask;
1989 struct ib_flow_gre_filter {
1990 __be16 c_ks_res0_ver;
1997 struct ib_flow_spec_gre {
2000 struct ib_flow_gre_filter val;
2001 struct ib_flow_gre_filter mask;
2004 struct ib_flow_mpls_filter {
2010 struct ib_flow_spec_mpls {
2013 struct ib_flow_mpls_filter val;
2014 struct ib_flow_mpls_filter mask;
2017 struct ib_flow_spec_action_tag {
2018 enum ib_flow_spec_type type;
2023 struct ib_flow_spec_action_drop {
2024 enum ib_flow_spec_type type;
2028 struct ib_flow_spec_action_handle {
2029 enum ib_flow_spec_type type;
2031 struct ib_flow_action *act;
2034 enum ib_counters_description {
2039 struct ib_flow_spec_action_count {
2040 enum ib_flow_spec_type type;
2042 struct ib_counters *counters;
2045 union ib_flow_spec {
2050 struct ib_flow_spec_eth eth;
2051 struct ib_flow_spec_ib ib;
2052 struct ib_flow_spec_ipv4 ipv4;
2053 struct ib_flow_spec_tcp_udp tcp_udp;
2054 struct ib_flow_spec_ipv6 ipv6;
2055 struct ib_flow_spec_tunnel tunnel;
2056 struct ib_flow_spec_esp esp;
2057 struct ib_flow_spec_gre gre;
2058 struct ib_flow_spec_mpls mpls;
2059 struct ib_flow_spec_action_tag flow_tag;
2060 struct ib_flow_spec_action_drop drop;
2061 struct ib_flow_spec_action_handle action;
2062 struct ib_flow_spec_action_count flow_count;
2065 struct ib_flow_attr {
2066 enum ib_flow_attr_type type;
2072 union ib_flow_spec flows[];
2077 struct ib_device *device;
2078 struct ib_uobject *uobject;
2081 enum ib_flow_action_type {
2082 IB_FLOW_ACTION_UNSPECIFIED,
2083 IB_FLOW_ACTION_ESP = 1,
2086 struct ib_flow_action_attrs_esp_keymats {
2087 enum ib_uverbs_flow_action_esp_keymat protocol;
2089 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2093 struct ib_flow_action_attrs_esp_replays {
2094 enum ib_uverbs_flow_action_esp_replay protocol;
2096 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2100 enum ib_flow_action_attrs_esp_flags {
2101 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2102 * This is done in order to share the same flags between user-space and
2103 * kernel and spare an unnecessary translation.
2107 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2108 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2111 struct ib_flow_spec_list {
2112 struct ib_flow_spec_list *next;
2113 union ib_flow_spec spec;
2116 struct ib_flow_action_attrs_esp {
2117 struct ib_flow_action_attrs_esp_keymats *keymat;
2118 struct ib_flow_action_attrs_esp_replays *replay;
2119 struct ib_flow_spec_list *encap;
2120 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2121 * Value of 0 is a valid value.
2127 /* Use enum ib_flow_action_attrs_esp_flags */
2129 u64 hard_limit_pkts;
2132 struct ib_flow_action {
2133 struct ib_device *device;
2134 struct ib_uobject *uobject;
2135 enum ib_flow_action_type type;
2142 enum ib_process_mad_flags {
2143 IB_MAD_IGNORE_MKEY = 1,
2144 IB_MAD_IGNORE_BKEY = 2,
2145 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2148 enum ib_mad_result {
2149 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2150 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2151 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2152 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2155 struct ib_port_cache {
2157 struct ib_pkey_cache *pkey;
2158 struct ib_gid_table *gid;
2160 enum ib_port_state port_state;
2163 struct ib_port_immutable {
2170 struct ib_port_data {
2171 struct ib_device *ib_dev;
2173 struct ib_port_immutable immutable;
2175 spinlock_t pkey_list_lock;
2176 struct list_head pkey_list;
2178 struct ib_port_cache cache;
2180 spinlock_t netdev_lock;
2181 struct net_device __rcu *netdev;
2182 struct hlist_node ndev_hash_link;
2183 struct rdma_port_counter port_counter;
2184 struct rdma_hw_stats *hw_stats;
2187 /* rdma netdev type - specifies protocol type */
2188 enum rdma_netdev_t {
2189 RDMA_NETDEV_OPA_VNIC,
2194 * struct rdma_netdev - rdma netdev
2195 * For cases where netstack interfacing is required.
2197 struct rdma_netdev {
2199 struct ib_device *hca;
2203 * cleanup function must be specified.
2204 * FIXME: This is only used for OPA_VNIC and that usage should be
2207 void (*free_rdma_netdev)(struct net_device *netdev);
2209 /* control functions */
2210 void (*set_id)(struct net_device *netdev, int id);
2212 int (*send)(struct net_device *dev, struct sk_buff *skb,
2213 struct ib_ah *address, u32 dqpn);
2215 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2216 union ib_gid *gid, u16 mlid,
2217 int set_qkey, u32 qkey);
2218 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2219 union ib_gid *gid, u16 mlid);
2222 struct rdma_netdev_alloc_params {
2228 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2229 struct net_device *netdev, void *param);
2232 struct ib_odp_counters {
2234 atomic64_t invalidations;
2237 struct ib_counters {
2238 struct ib_device *device;
2239 struct ib_uobject *uobject;
2240 /* num of objects attached */
2244 struct ib_counters_read_attr {
2247 u32 flags; /* use enum ib_read_counters_flags */
2250 struct uverbs_attr_bundle;
2252 struct iw_cm_conn_param;
2254 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2255 .size_##ib_struct = \
2256 (sizeof(struct drv_struct) + \
2257 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2258 BUILD_BUG_ON_ZERO( \
2259 !__same_type(((struct drv_struct *)NULL)->member, \
2262 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2263 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2265 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2266 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2268 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2270 struct rdma_user_mmap_entry {
2272 struct ib_ucontext *ucontext;
2273 unsigned long start_pgoff;
2275 bool driver_removed;
2278 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2280 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2282 return (u64)entry->start_pgoff << PAGE_SHIFT;
2286 * struct ib_device_ops - InfiniBand device operations
2287 * This structure defines all the InfiniBand device operations, providers will
2288 * need to define the supported operations, otherwise they will be set to null.
2290 struct ib_device_ops {
2291 struct module *owner;
2292 enum rdma_driver_id driver_id;
2294 unsigned int uverbs_no_driver_id_binding:1;
2296 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2297 const struct ib_send_wr **bad_send_wr);
2298 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2299 const struct ib_recv_wr **bad_recv_wr);
2300 void (*drain_rq)(struct ib_qp *qp);
2301 void (*drain_sq)(struct ib_qp *qp);
2302 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2303 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2304 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2305 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2306 int (*post_srq_recv)(struct ib_srq *srq,
2307 const struct ib_recv_wr *recv_wr,
2308 const struct ib_recv_wr **bad_recv_wr);
2309 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2310 u8 port_num, const struct ib_wc *in_wc,
2311 const struct ib_grh *in_grh,
2312 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2313 size_t *out_mad_size, u16 *out_mad_pkey_index);
2314 int (*query_device)(struct ib_device *device,
2315 struct ib_device_attr *device_attr,
2316 struct ib_udata *udata);
2317 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2318 struct ib_device_modify *device_modify);
2319 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2320 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2322 int (*query_port)(struct ib_device *device, u8 port_num,
2323 struct ib_port_attr *port_attr);
2324 int (*modify_port)(struct ib_device *device, u8 port_num,
2325 int port_modify_mask,
2326 struct ib_port_modify *port_modify);
2328 * The following mandatory functions are used only at device
2329 * registration. Keep functions such as these at the end of this
2330 * structure to avoid cache line misses when accessing struct ib_device
2333 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2334 struct ib_port_immutable *immutable);
2335 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2338 * When calling get_netdev, the HW vendor's driver should return the
2339 * net device of device @device at port @port_num or NULL if such
2340 * a net device doesn't exist. The vendor driver should call dev_hold
2341 * on this net device. The HW vendor's device driver must guarantee
2342 * that this function returns NULL before the net device has finished
2343 * NETDEV_UNREGISTER state.
2345 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2347 * rdma netdev operation
2349 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2350 * must return -EOPNOTSUPP if it doesn't support the specified type.
2352 struct net_device *(*alloc_rdma_netdev)(
2353 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2354 const char *name, unsigned char name_assign_type,
2355 void (*setup)(struct net_device *));
2357 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2358 enum rdma_netdev_t type,
2359 struct rdma_netdev_alloc_params *params);
2361 * query_gid should be return GID value for @device, when @port_num
2362 * link layer is either IB or iWarp. It is no-op if @port_num port
2363 * is RoCE link layer.
2365 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2368 * When calling add_gid, the HW vendor's driver should add the gid
2369 * of device of port at gid index available at @attr. Meta-info of
2370 * that gid (for example, the network device related to this gid) is
2371 * available at @attr. @context allows the HW vendor driver to store
2372 * extra information together with a GID entry. The HW vendor driver may
2373 * allocate memory to contain this information and store it in @context
2374 * when a new GID entry is written to. Params are consistent until the
2375 * next call of add_gid or delete_gid. The function should return 0 on
2376 * success or error otherwise. The function could be called
2377 * concurrently for different ports. This function is only called when
2378 * roce_gid_table is used.
2380 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2382 * When calling del_gid, the HW vendor's driver should delete the
2383 * gid of device @device at gid index gid_index of port port_num
2384 * available in @attr.
2385 * Upon the deletion of a GID entry, the HW vendor must free any
2386 * allocated memory. The caller will clear @context afterwards.
2387 * This function is only called when roce_gid_table is used.
2389 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2390 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2392 int (*alloc_ucontext)(struct ib_ucontext *context,
2393 struct ib_udata *udata);
2394 void (*dealloc_ucontext)(struct ib_ucontext *context);
2395 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2397 * This will be called once refcount of an entry in mmap_xa reaches
2398 * zero. The type of the memory that was mapped may differ between
2399 * entries and is opaque to the rdma_user_mmap interface.
2400 * Therefore needs to be implemented by the driver in mmap_free.
2402 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2403 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2404 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2405 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2406 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2407 u32 flags, struct ib_udata *udata);
2408 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2409 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2410 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2411 int (*create_srq)(struct ib_srq *srq,
2412 struct ib_srq_init_attr *srq_init_attr,
2413 struct ib_udata *udata);
2414 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2415 enum ib_srq_attr_mask srq_attr_mask,
2416 struct ib_udata *udata);
2417 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2418 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2419 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2420 struct ib_qp_init_attr *qp_init_attr,
2421 struct ib_udata *udata);
2422 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2423 int qp_attr_mask, struct ib_udata *udata);
2424 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2425 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2426 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2427 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2428 struct ib_udata *udata);
2429 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2430 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2431 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2432 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2433 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2434 u64 virt_addr, int mr_access_flags,
2435 struct ib_udata *udata);
2436 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2437 u64 virt_addr, int mr_access_flags,
2438 struct ib_pd *pd, struct ib_udata *udata);
2439 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2440 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2441 u32 max_num_sg, struct ib_udata *udata);
2442 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2443 u32 max_num_data_sg,
2444 u32 max_num_meta_sg);
2445 int (*advise_mr)(struct ib_pd *pd,
2446 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2447 struct ib_sge *sg_list, u32 num_sge,
2448 struct uverbs_attr_bundle *attrs);
2449 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2450 unsigned int *sg_offset);
2451 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2452 struct ib_mr_status *mr_status);
2453 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2454 struct ib_udata *udata);
2455 int (*dealloc_mw)(struct ib_mw *mw);
2456 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2457 struct ib_fmr_attr *fmr_attr);
2458 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2460 int (*unmap_fmr)(struct list_head *fmr_list);
2461 int (*dealloc_fmr)(struct ib_fmr *fmr);
2462 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2463 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2464 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2465 struct ib_udata *udata);
2466 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2467 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2468 struct ib_flow_attr *flow_attr,
2469 int domain, struct ib_udata *udata);
2470 int (*destroy_flow)(struct ib_flow *flow_id);
2471 struct ib_flow_action *(*create_flow_action_esp)(
2472 struct ib_device *device,
2473 const struct ib_flow_action_attrs_esp *attr,
2474 struct uverbs_attr_bundle *attrs);
2475 int (*destroy_flow_action)(struct ib_flow_action *action);
2476 int (*modify_flow_action_esp)(
2477 struct ib_flow_action *action,
2478 const struct ib_flow_action_attrs_esp *attr,
2479 struct uverbs_attr_bundle *attrs);
2480 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2482 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2483 struct ifla_vf_info *ivf);
2484 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2485 struct ifla_vf_stats *stats);
2486 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2487 struct ifla_vf_guid *node_guid,
2488 struct ifla_vf_guid *port_guid);
2489 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2491 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2492 struct ib_wq_init_attr *init_attr,
2493 struct ib_udata *udata);
2494 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2495 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2496 u32 wq_attr_mask, struct ib_udata *udata);
2497 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2498 struct ib_device *device,
2499 struct ib_rwq_ind_table_init_attr *init_attr,
2500 struct ib_udata *udata);
2501 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2502 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2503 struct ib_ucontext *context,
2504 struct ib_dm_alloc_attr *attr,
2505 struct uverbs_attr_bundle *attrs);
2506 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2507 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2508 struct ib_dm_mr_attr *attr,
2509 struct uverbs_attr_bundle *attrs);
2510 struct ib_counters *(*create_counters)(
2511 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2512 int (*destroy_counters)(struct ib_counters *counters);
2513 int (*read_counters)(struct ib_counters *counters,
2514 struct ib_counters_read_attr *counters_read_attr,
2515 struct uverbs_attr_bundle *attrs);
2516 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2517 int data_sg_nents, unsigned int *data_sg_offset,
2518 struct scatterlist *meta_sg, int meta_sg_nents,
2519 unsigned int *meta_sg_offset);
2522 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2523 * driver initialized data. The struct is kfree()'ed by the sysfs
2524 * core when the device is removed. A lifespan of -1 in the return
2525 * struct tells the core to set a default lifespan.
2527 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2530 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2531 * @index - The index in the value array we wish to have updated, or
2532 * num_counters if we want all stats updated
2534 * < 0 - Error, no counters updated
2535 * index - Updated the single counter pointed to by index
2536 * num_counters - Updated all counters (will reset the timestamp
2537 * and prevent further calls for lifespan milliseconds)
2538 * Drivers are allowed to update all counters in leiu of just the
2539 * one given in index at their option
2541 int (*get_hw_stats)(struct ib_device *device,
2542 struct rdma_hw_stats *stats, u8 port, int index);
2544 * This function is called once for each port when a ib device is
2547 int (*init_port)(struct ib_device *device, u8 port_num,
2548 struct kobject *port_sysfs);
2550 * Allows rdma drivers to add their own restrack attributes.
2552 int (*fill_res_entry)(struct sk_buff *msg,
2553 struct rdma_restrack_entry *entry);
2555 /* Device lifecycle callbacks */
2557 * Called after the device becomes registered, before clients are
2560 int (*enable_driver)(struct ib_device *dev);
2562 * This is called as part of ib_dealloc_device().
2564 void (*dealloc_driver)(struct ib_device *dev);
2566 /* iWarp CM callbacks */
2567 void (*iw_add_ref)(struct ib_qp *qp);
2568 void (*iw_rem_ref)(struct ib_qp *qp);
2569 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2570 int (*iw_connect)(struct iw_cm_id *cm_id,
2571 struct iw_cm_conn_param *conn_param);
2572 int (*iw_accept)(struct iw_cm_id *cm_id,
2573 struct iw_cm_conn_param *conn_param);
2574 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2576 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2577 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2579 * counter_bind_qp - Bind a QP to a counter.
2580 * @counter - The counter to be bound. If counter->id is zero then
2581 * the driver needs to allocate a new counter and set counter->id
2583 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2585 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2586 * counter and bind it onto the default one
2588 int (*counter_unbind_qp)(struct ib_qp *qp);
2590 * counter_dealloc -De-allocate the hw counter
2592 int (*counter_dealloc)(struct rdma_counter *counter);
2594 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2595 * the driver initialized data.
2597 struct rdma_hw_stats *(*counter_alloc_stats)(
2598 struct rdma_counter *counter);
2600 * counter_update_stats - Query the stats value of this counter
2602 int (*counter_update_stats)(struct rdma_counter *counter);
2605 * Allows rdma drivers to add their own restrack attributes
2606 * dumped via 'rdma stat' iproute2 command.
2608 int (*fill_stat_entry)(struct sk_buff *msg,
2609 struct rdma_restrack_entry *entry);
2611 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2612 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2613 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2614 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2615 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2618 struct ib_core_device {
2619 /* device must be the first element in structure until,
2620 * union of ib_core_device and device exists in ib_device.
2623 possible_net_t rdma_net;
2624 struct kobject *ports_kobj;
2625 struct list_head port_list;
2626 struct ib_device *owner; /* reach back to owner ib_device */
2629 struct rdma_restrack_root;
2631 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2632 struct device *dma_device;
2633 struct ib_device_ops ops;
2634 char name[IB_DEVICE_NAME_MAX];
2635 struct rcu_head rcu_head;
2637 struct list_head event_handler_list;
2638 /* Protects event_handler_list */
2639 struct rw_semaphore event_handler_rwsem;
2641 /* Protects QP's event_handler calls and open_qp list */
2642 spinlock_t qp_open_list_lock;
2644 struct rw_semaphore client_data_rwsem;
2645 struct xarray client_data;
2646 struct mutex unregistration_lock;
2648 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2649 rwlock_t cache_lock;
2651 * port_data is indexed by port number
2653 struct ib_port_data *port_data;
2655 int num_comp_vectors;
2659 struct ib_core_device coredev;
2662 /* First group for device attributes,
2663 * Second group for driver provided attributes (optional).
2664 * It is NULL terminated array.
2666 const struct attribute_group *groups[3];
2668 u64 uverbs_cmd_mask;
2669 u64 uverbs_ex_cmd_mask;
2671 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2675 /* Indicates kernel verbs support, should not be used in drivers */
2676 u16 kverbs_provider:1;
2677 /* CQ adaptive moderation (RDMA DIM) */
2681 struct ib_device_attr attrs;
2682 struct attribute_group *hw_stats_ag;
2683 struct rdma_hw_stats *hw_stats;
2685 #ifdef CONFIG_CGROUP_RDMA
2686 struct rdmacg_device cg_device;
2690 struct rdma_restrack_root *res;
2692 const struct uapi_definition *driver_def;
2695 * Positive refcount indicates that the device is currently
2696 * registered and cannot be unregistered.
2698 refcount_t refcount;
2699 struct completion unreg_completion;
2700 struct work_struct unregistration_work;
2702 const struct rdma_link_ops *link_ops;
2704 /* Protects compat_devs xarray modifications */
2705 struct mutex compat_devs_mutex;
2706 /* Maintains compat devices for each net namespace */
2707 struct xarray compat_devs;
2709 /* Used by iWarp CM */
2710 char iw_ifname[IFNAMSIZ];
2711 u32 iw_driver_flags;
2714 struct ib_client_nl_info;
2717 void (*add) (struct ib_device *);
2718 void (*remove)(struct ib_device *, void *client_data);
2719 void (*rename)(struct ib_device *dev, void *client_data);
2720 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2721 struct ib_client_nl_info *res);
2722 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2724 /* Returns the net_dev belonging to this ib_client and matching the
2726 * @dev: An RDMA device that the net_dev use for communication.
2727 * @port: A physical port number on the RDMA device.
2728 * @pkey: P_Key that the net_dev uses if applicable.
2729 * @gid: A GID that the net_dev uses to communicate.
2730 * @addr: An IP address the net_dev is configured with.
2731 * @client_data: The device's client data set by ib_set_client_data().
2733 * An ib_client that implements a net_dev on top of RDMA devices
2734 * (such as IP over IB) should implement this callback, allowing the
2735 * rdma_cm module to find the right net_dev for a given request.
2737 * The caller is responsible for calling dev_put on the returned
2739 struct net_device *(*get_net_dev_by_params)(
2740 struct ib_device *dev,
2743 const union ib_gid *gid,
2744 const struct sockaddr *addr,
2748 struct completion uses_zero;
2751 /* kverbs are not required by the client */
2756 * IB block DMA iterator
2758 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2759 * to a HW supported page size.
2761 struct ib_block_iter {
2762 /* internal states */
2763 struct scatterlist *__sg; /* sg holding the current aligned block */
2764 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2765 unsigned int __sg_nents; /* number of SG entries */
2766 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2767 unsigned int __pg_bit; /* alignment of current block */
2770 struct ib_device *_ib_alloc_device(size_t size);
2771 #define ib_alloc_device(drv_struct, member) \
2772 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2773 BUILD_BUG_ON_ZERO(offsetof( \
2774 struct drv_struct, member))), \
2775 struct drv_struct, member)
2777 void ib_dealloc_device(struct ib_device *device);
2779 void ib_get_device_fw_str(struct ib_device *device, char *str);
2781 int ib_register_device(struct ib_device *device, const char *name);
2782 void ib_unregister_device(struct ib_device *device);
2783 void ib_unregister_driver(enum rdma_driver_id driver_id);
2784 void ib_unregister_device_and_put(struct ib_device *device);
2785 void ib_unregister_device_queued(struct ib_device *ib_dev);
2787 int ib_register_client (struct ib_client *client);
2788 void ib_unregister_client(struct ib_client *client);
2790 void __rdma_block_iter_start(struct ib_block_iter *biter,
2791 struct scatterlist *sglist,
2793 unsigned long pgsz);
2794 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2797 * rdma_block_iter_dma_address - get the aligned dma address of the current
2798 * block held by the block iterator.
2799 * @biter: block iterator holding the memory block
2801 static inline dma_addr_t
2802 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2804 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2808 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2809 * @sglist: sglist to iterate over
2810 * @biter: block iterator holding the memory block
2811 * @nents: maximum number of sg entries to iterate over
2812 * @pgsz: best HW supported page size to use
2814 * Callers may use rdma_block_iter_dma_address() to get each
2815 * blocks aligned DMA address.
2817 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2818 for (__rdma_block_iter_start(biter, sglist, nents, \
2820 __rdma_block_iter_next(biter);)
2823 * ib_get_client_data - Get IB client context
2824 * @device:Device to get context for
2825 * @client:Client to get context for
2827 * ib_get_client_data() returns the client context data set with
2828 * ib_set_client_data(). This can only be called while the client is
2829 * registered to the device, once the ib_client remove() callback returns this
2832 static inline void *ib_get_client_data(struct ib_device *device,
2833 struct ib_client *client)
2835 return xa_load(&device->client_data, client->client_id);
2837 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2839 void ib_set_device_ops(struct ib_device *device,
2840 const struct ib_device_ops *ops);
2842 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2843 unsigned long pfn, unsigned long size, pgprot_t prot,
2844 struct rdma_user_mmap_entry *entry);
2845 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2846 struct rdma_user_mmap_entry *entry,
2848 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2849 struct rdma_user_mmap_entry *entry,
2850 size_t length, u32 min_pgoff,
2853 struct rdma_user_mmap_entry *
2854 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2855 unsigned long pgoff);
2856 struct rdma_user_mmap_entry *
2857 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2858 struct vm_area_struct *vma);
2859 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2861 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2863 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2865 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2868 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2870 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2873 static inline bool ib_is_buffer_cleared(const void __user *p,
2879 if (len > USHRT_MAX)
2882 buf = memdup_user(p, len);
2886 ret = !memchr_inv(buf, 0, len);
2891 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2895 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2899 * ib_is_destroy_retryable - Check whether the uobject destruction
2901 * @ret: The initial destruction return code
2902 * @why: remove reason
2903 * @uobj: The uobject that is destroyed
2905 * This function is a helper function that IB layer and low-level drivers
2906 * can use to consider whether the destruction of the given uobject is
2908 * It checks the original return code, if it wasn't success the destruction
2909 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2910 * the remove reason. (i.e. why).
2911 * Must be called with the object locked for destroy.
2913 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2914 struct ib_uobject *uobj)
2916 return ret && (why == RDMA_REMOVE_DESTROY ||
2917 uobj->context->cleanup_retryable);
2921 * ib_destroy_usecnt - Called during destruction to check the usecnt
2922 * @usecnt: The usecnt atomic
2923 * @why: remove reason
2924 * @uobj: The uobject that is destroyed
2926 * Non-zero usecnts will block destruction unless destruction was triggered by
2927 * a ucontext cleanup.
2929 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2930 enum rdma_remove_reason why,
2931 struct ib_uobject *uobj)
2933 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2939 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2940 * contains all required attributes and no attributes not allowed for
2941 * the given QP state transition.
2942 * @cur_state: Current QP state
2943 * @next_state: Next QP state
2945 * @mask: Mask of supplied QP attributes
2947 * This function is a helper function that a low-level driver's
2948 * modify_qp method can use to validate the consumer's input. It
2949 * checks that cur_state and next_state are valid QP states, that a
2950 * transition from cur_state to next_state is allowed by the IB spec,
2951 * and that the attribute mask supplied is allowed for the transition.
2953 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2954 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2956 void ib_register_event_handler(struct ib_event_handler *event_handler);
2957 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2958 void ib_dispatch_event(const struct ib_event *event);
2960 int ib_query_port(struct ib_device *device,
2961 u8 port_num, struct ib_port_attr *port_attr);
2963 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2967 * rdma_cap_ib_switch - Check if the device is IB switch
2968 * @device: Device to check
2970 * Device driver is responsible for setting is_switch bit on
2971 * in ib_device structure at init time.
2973 * Return: true if the device is IB switch.
2975 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2977 return device->is_switch;
2981 * rdma_start_port - Return the first valid port number for the device
2984 * @device: Device to be checked
2986 * Return start port number
2988 static inline u8 rdma_start_port(const struct ib_device *device)
2990 return rdma_cap_ib_switch(device) ? 0 : 1;
2994 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2995 * @device - The struct ib_device * to iterate over
2996 * @iter - The unsigned int to store the port number
2998 #define rdma_for_each_port(device, iter) \
2999 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3000 unsigned int, iter))); \
3001 iter <= rdma_end_port(device); (iter)++)
3004 * rdma_end_port - Return the last valid port number for the device
3007 * @device: Device to be checked
3009 * Return last port number
3011 static inline u8 rdma_end_port(const struct ib_device *device)
3013 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3016 static inline int rdma_is_port_valid(const struct ib_device *device,
3019 return (port >= rdma_start_port(device) &&
3020 port <= rdma_end_port(device));
3023 static inline bool rdma_is_grh_required(const struct ib_device *device,
3026 return device->port_data[port_num].immutable.core_cap_flags &
3027 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3030 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3032 return device->port_data[port_num].immutable.core_cap_flags &
3033 RDMA_CORE_CAP_PROT_IB;
3036 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3038 return device->port_data[port_num].immutable.core_cap_flags &
3039 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3042 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3044 return device->port_data[port_num].immutable.core_cap_flags &
3045 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3048 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3050 return device->port_data[port_num].immutable.core_cap_flags &
3051 RDMA_CORE_CAP_PROT_ROCE;
3054 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3056 return device->port_data[port_num].immutable.core_cap_flags &
3057 RDMA_CORE_CAP_PROT_IWARP;
3060 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3062 return rdma_protocol_ib(device, port_num) ||
3063 rdma_protocol_roce(device, port_num);
3066 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3068 return device->port_data[port_num].immutable.core_cap_flags &
3069 RDMA_CORE_CAP_PROT_RAW_PACKET;
3072 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3074 return device->port_data[port_num].immutable.core_cap_flags &
3075 RDMA_CORE_CAP_PROT_USNIC;
3079 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3080 * Management Datagrams.
3081 * @device: Device to check
3082 * @port_num: Port number to check
3084 * Management Datagrams (MAD) are a required part of the InfiniBand
3085 * specification and are supported on all InfiniBand devices. A slightly
3086 * extended version are also supported on OPA interfaces.
3088 * Return: true if the port supports sending/receiving of MAD packets.
3090 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3092 return device->port_data[port_num].immutable.core_cap_flags &
3093 RDMA_CORE_CAP_IB_MAD;
3097 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3098 * Management Datagrams.
3099 * @device: Device to check
3100 * @port_num: Port number to check
3102 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3103 * datagrams with their own versions. These OPA MADs share many but not all of
3104 * the characteristics of InfiniBand MADs.
3106 * OPA MADs differ in the following ways:
3108 * 1) MADs are variable size up to 2K
3109 * IBTA defined MADs remain fixed at 256 bytes
3110 * 2) OPA SMPs must carry valid PKeys
3111 * 3) OPA SMP packets are a different format
3113 * Return: true if the port supports OPA MAD packet formats.
3115 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3117 return device->port_data[port_num].immutable.core_cap_flags &
3118 RDMA_CORE_CAP_OPA_MAD;
3122 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3123 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3124 * @device: Device to check
3125 * @port_num: Port number to check
3127 * Each InfiniBand node is required to provide a Subnet Management Agent
3128 * that the subnet manager can access. Prior to the fabric being fully
3129 * configured by the subnet manager, the SMA is accessed via a well known
3130 * interface called the Subnet Management Interface (SMI). This interface
3131 * uses directed route packets to communicate with the SM to get around the
3132 * chicken and egg problem of the SM needing to know what's on the fabric
3133 * in order to configure the fabric, and needing to configure the fabric in
3134 * order to send packets to the devices on the fabric. These directed
3135 * route packets do not need the fabric fully configured in order to reach
3136 * their destination. The SMI is the only method allowed to send
3137 * directed route packets on an InfiniBand fabric.
3139 * Return: true if the port provides an SMI.
3141 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3143 return device->port_data[port_num].immutable.core_cap_flags &
3144 RDMA_CORE_CAP_IB_SMI;
3148 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3149 * Communication Manager.
3150 * @device: Device to check
3151 * @port_num: Port number to check
3153 * The InfiniBand Communication Manager is one of many pre-defined General
3154 * Service Agents (GSA) that are accessed via the General Service
3155 * Interface (GSI). It's role is to facilitate establishment of connections
3156 * between nodes as well as other management related tasks for established
3159 * Return: true if the port supports an IB CM (this does not guarantee that
3160 * a CM is actually running however).
3162 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3164 return device->port_data[port_num].immutable.core_cap_flags &
3165 RDMA_CORE_CAP_IB_CM;
3169 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3170 * Communication Manager.
3171 * @device: Device to check
3172 * @port_num: Port number to check
3174 * Similar to above, but specific to iWARP connections which have a different
3175 * managment protocol than InfiniBand.
3177 * Return: true if the port supports an iWARP CM (this does not guarantee that
3178 * a CM is actually running however).
3180 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3182 return device->port_data[port_num].immutable.core_cap_flags &
3183 RDMA_CORE_CAP_IW_CM;
3187 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3188 * Subnet Administration.
3189 * @device: Device to check
3190 * @port_num: Port number to check
3192 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3193 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3194 * fabrics, devices should resolve routes to other hosts by contacting the
3195 * SA to query the proper route.
3197 * Return: true if the port should act as a client to the fabric Subnet
3198 * Administration interface. This does not imply that the SA service is
3201 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3203 return device->port_data[port_num].immutable.core_cap_flags &
3204 RDMA_CORE_CAP_IB_SA;
3208 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3210 * @device: Device to check
3211 * @port_num: Port number to check
3213 * InfiniBand multicast registration is more complex than normal IPv4 or
3214 * IPv6 multicast registration. Each Host Channel Adapter must register
3215 * with the Subnet Manager when it wishes to join a multicast group. It
3216 * should do so only once regardless of how many queue pairs it subscribes
3217 * to this group. And it should leave the group only after all queue pairs
3218 * attached to the group have been detached.
3220 * Return: true if the port must undertake the additional adminstrative
3221 * overhead of registering/unregistering with the SM and tracking of the
3222 * total number of queue pairs attached to the multicast group.
3224 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3226 return rdma_cap_ib_sa(device, port_num);
3230 * rdma_cap_af_ib - Check if the port of device has the capability
3231 * Native Infiniband Address.
3232 * @device: Device to check
3233 * @port_num: Port number to check
3235 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3236 * GID. RoCE uses a different mechanism, but still generates a GID via
3237 * a prescribed mechanism and port specific data.
3239 * Return: true if the port uses a GID address to identify devices on the
3242 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3244 return device->port_data[port_num].immutable.core_cap_flags &
3245 RDMA_CORE_CAP_AF_IB;
3249 * rdma_cap_eth_ah - Check if the port of device has the capability
3250 * Ethernet Address Handle.
3251 * @device: Device to check
3252 * @port_num: Port number to check
3254 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3255 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3256 * port. Normally, packet headers are generated by the sending host
3257 * adapter, but when sending connectionless datagrams, we must manually
3258 * inject the proper headers for the fabric we are communicating over.
3260 * Return: true if we are running as a RoCE port and must force the
3261 * addition of a Global Route Header built from our Ethernet Address
3262 * Handle into our header list for connectionless packets.
3264 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3266 return device->port_data[port_num].immutable.core_cap_flags &
3267 RDMA_CORE_CAP_ETH_AH;
3271 * rdma_cap_opa_ah - Check if the port of device supports
3272 * OPA Address handles
3273 * @device: Device to check
3274 * @port_num: Port number to check
3276 * Return: true if we are running on an OPA device which supports
3277 * the extended OPA addressing.
3279 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3281 return (device->port_data[port_num].immutable.core_cap_flags &
3282 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3286 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3289 * @port_num: Port number
3291 * This MAD size includes the MAD headers and MAD payload. No other headers
3294 * Return the max MAD size required by the Port. Will return 0 if the port
3295 * does not support MADs
3297 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3299 return device->port_data[port_num].immutable.max_mad_size;
3303 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3304 * @device: Device to check
3305 * @port_num: Port number to check
3307 * RoCE GID table mechanism manages the various GIDs for a device.
3309 * NOTE: if allocating the port's GID table has failed, this call will still
3310 * return true, but any RoCE GID table API will fail.
3312 * Return: true if the port uses RoCE GID table mechanism in order to manage
3315 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3318 return rdma_protocol_roce(device, port_num) &&
3319 device->ops.add_gid && device->ops.del_gid;
3323 * Check if the device supports READ W/ INVALIDATE.
3325 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3328 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3329 * has support for it yet.
3331 return rdma_protocol_iwarp(dev, port_num);
3335 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3338 * @pgsz_bitmap: bitmap of HW supported page sizes
3340 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3341 unsigned long pgsz_bitmap)
3343 unsigned long align;
3346 align = addr & -addr;
3348 /* Find page bit such that addr is aligned to the highest supported
3351 pgsz = pgsz_bitmap & ~(-align << 1);
3353 return __ffs(pgsz_bitmap);
3358 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3360 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3361 struct ifla_vf_info *info);
3362 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3363 struct ifla_vf_stats *stats);
3364 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3365 struct ifla_vf_guid *node_guid,
3366 struct ifla_vf_guid *port_guid);
3367 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3370 int ib_query_pkey(struct ib_device *device,
3371 u8 port_num, u16 index, u16 *pkey);
3373 int ib_modify_device(struct ib_device *device,
3374 int device_modify_mask,
3375 struct ib_device_modify *device_modify);
3377 int ib_modify_port(struct ib_device *device,
3378 u8 port_num, int port_modify_mask,
3379 struct ib_port_modify *port_modify);
3381 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3382 u8 *port_num, u16 *index);
3384 int ib_find_pkey(struct ib_device *device,
3385 u8 port_num, u16 pkey, u16 *index);
3389 * Create a memory registration for all memory in the system and place
3390 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3391 * ULPs to avoid the overhead of dynamic MRs.
3393 * This flag is generally considered unsafe and must only be used in
3394 * extremly trusted environments. Every use of it will log a warning
3395 * in the kernel log.
3397 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3400 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3401 const char *caller);
3403 #define ib_alloc_pd(device, flags) \
3404 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3407 * ib_dealloc_pd_user - Deallocate kernel/user PD
3408 * @pd: The protection domain
3409 * @udata: Valid user data or NULL for kernel objects
3411 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3414 * ib_dealloc_pd - Deallocate kernel PD
3415 * @pd: The protection domain
3417 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3419 static inline void ib_dealloc_pd(struct ib_pd *pd)
3421 ib_dealloc_pd_user(pd, NULL);
3424 enum rdma_create_ah_flags {
3425 /* In a sleepable context */
3426 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3430 * rdma_create_ah - Creates an address handle for the given address vector.
3431 * @pd: The protection domain associated with the address handle.
3432 * @ah_attr: The attributes of the address vector.
3433 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3435 * The address handle is used to reference a local or global destination
3436 * in all UD QP post sends.
3438 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3442 * rdma_create_user_ah - Creates an address handle for the given address vector.
3443 * It resolves destination mac address for ah attribute of RoCE type.
3444 * @pd: The protection domain associated with the address handle.
3445 * @ah_attr: The attributes of the address vector.
3446 * @udata: pointer to user's input output buffer information need by
3449 * It returns 0 on success and returns appropriate error code on error.
3450 * The address handle is used to reference a local or global destination
3451 * in all UD QP post sends.
3453 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3454 struct rdma_ah_attr *ah_attr,
3455 struct ib_udata *udata);
3457 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3459 * @hdr: the L3 header to parse
3460 * @net_type: type of header to parse
3461 * @sgid: place to store source gid
3462 * @dgid: place to store destination gid
3464 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3465 enum rdma_network_type net_type,
3466 union ib_gid *sgid, union ib_gid *dgid);
3469 * ib_get_rdma_header_version - Get the header version
3470 * @hdr: the L3 header to parse
3472 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3475 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3477 * @device: Device on which the received message arrived.
3478 * @port_num: Port on which the received message arrived.
3479 * @wc: Work completion associated with the received message.
3480 * @grh: References the received global route header. This parameter is
3481 * ignored unless the work completion indicates that the GRH is valid.
3482 * @ah_attr: Returned attributes that can be used when creating an address
3483 * handle for replying to the message.
3484 * When ib_init_ah_attr_from_wc() returns success,
3485 * (a) for IB link layer it optionally contains a reference to SGID attribute
3486 * when GRH is present for IB link layer.
3487 * (b) for RoCE link layer it contains a reference to SGID attribute.
3488 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3489 * attributes which are initialized using ib_init_ah_attr_from_wc().
3492 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3493 const struct ib_wc *wc, const struct ib_grh *grh,
3494 struct rdma_ah_attr *ah_attr);
3497 * ib_create_ah_from_wc - Creates an address handle associated with the
3498 * sender of the specified work completion.
3499 * @pd: The protection domain associated with the address handle.
3500 * @wc: Work completion information associated with a received message.
3501 * @grh: References the received global route header. This parameter is
3502 * ignored unless the work completion indicates that the GRH is valid.
3503 * @port_num: The outbound port number to associate with the address.
3505 * The address handle is used to reference a local or global destination
3506 * in all UD QP post sends.
3508 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3509 const struct ib_grh *grh, u8 port_num);
3512 * rdma_modify_ah - Modifies the address vector associated with an address
3514 * @ah: The address handle to modify.
3515 * @ah_attr: The new address vector attributes to associate with the
3518 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3521 * rdma_query_ah - Queries the address vector associated with an address
3523 * @ah: The address handle to query.
3524 * @ah_attr: The address vector attributes associated with the address
3527 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3529 enum rdma_destroy_ah_flags {
3530 /* In a sleepable context */
3531 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3535 * rdma_destroy_ah_user - Destroys an address handle.
3536 * @ah: The address handle to destroy.
3537 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3538 * @udata: Valid user data or NULL for kernel objects
3540 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3543 * rdma_destroy_ah - Destroys an kernel address handle.
3544 * @ah: The address handle to destroy.
3545 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3547 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3549 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3551 return rdma_destroy_ah_user(ah, flags, NULL);
3555 * ib_create_srq - Creates a SRQ associated with the specified protection
3557 * @pd: The protection domain associated with the SRQ.
3558 * @srq_init_attr: A list of initial attributes required to create the
3559 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3560 * the actual capabilities of the created SRQ.
3562 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3563 * requested size of the SRQ, and set to the actual values allocated
3564 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3565 * will always be at least as large as the requested values.
3567 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3568 struct ib_srq_init_attr *srq_init_attr);
3571 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3572 * @srq: The SRQ to modify.
3573 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3574 * the current values of selected SRQ attributes are returned.
3575 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3576 * are being modified.
3578 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3579 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3580 * the number of receives queued drops below the limit.
3582 int ib_modify_srq(struct ib_srq *srq,
3583 struct ib_srq_attr *srq_attr,
3584 enum ib_srq_attr_mask srq_attr_mask);
3587 * ib_query_srq - Returns the attribute list and current values for the
3589 * @srq: The SRQ to query.
3590 * @srq_attr: The attributes of the specified SRQ.
3592 int ib_query_srq(struct ib_srq *srq,
3593 struct ib_srq_attr *srq_attr);
3596 * ib_destroy_srq_user - Destroys the specified SRQ.
3597 * @srq: The SRQ to destroy.
3598 * @udata: Valid user data or NULL for kernel objects
3600 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3603 * ib_destroy_srq - Destroys the specified kernel SRQ.
3604 * @srq: The SRQ to destroy.
3606 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3608 static inline int ib_destroy_srq(struct ib_srq *srq)
3610 return ib_destroy_srq_user(srq, NULL);
3614 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3615 * @srq: The SRQ to post the work request on.
3616 * @recv_wr: A list of work requests to post on the receive queue.
3617 * @bad_recv_wr: On an immediate failure, this parameter will reference
3618 * the work request that failed to be posted on the QP.
3620 static inline int ib_post_srq_recv(struct ib_srq *srq,
3621 const struct ib_recv_wr *recv_wr,
3622 const struct ib_recv_wr **bad_recv_wr)
3624 const struct ib_recv_wr *dummy;
3626 return srq->device->ops.post_srq_recv(srq, recv_wr,
3627 bad_recv_wr ? : &dummy);
3630 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3631 struct ib_qp_init_attr *qp_init_attr);
3634 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3635 * @qp: The QP to modify.
3636 * @attr: On input, specifies the QP attributes to modify. On output,
3637 * the current values of selected QP attributes are returned.
3638 * @attr_mask: A bit-mask used to specify which attributes of the QP
3639 * are being modified.
3640 * @udata: pointer to user's input output buffer information
3641 * are being modified.
3642 * It returns 0 on success and returns appropriate error code on error.
3644 int ib_modify_qp_with_udata(struct ib_qp *qp,
3645 struct ib_qp_attr *attr,
3647 struct ib_udata *udata);
3650 * ib_modify_qp - Modifies the attributes for the specified QP and then
3651 * transitions the QP to the given state.
3652 * @qp: The QP to modify.
3653 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3654 * the current values of selected QP attributes are returned.
3655 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3656 * are being modified.
3658 int ib_modify_qp(struct ib_qp *qp,
3659 struct ib_qp_attr *qp_attr,
3663 * ib_query_qp - Returns the attribute list and current values for the
3665 * @qp: The QP to query.
3666 * @qp_attr: The attributes of the specified QP.
3667 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3668 * @qp_init_attr: Additional attributes of the selected QP.
3670 * The qp_attr_mask may be used to limit the query to gathering only the
3671 * selected attributes.
3673 int ib_query_qp(struct ib_qp *qp,
3674 struct ib_qp_attr *qp_attr,
3676 struct ib_qp_init_attr *qp_init_attr);
3679 * ib_destroy_qp - Destroys the specified QP.
3680 * @qp: The QP to destroy.
3681 * @udata: Valid udata or NULL for kernel objects
3683 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3686 * ib_destroy_qp - Destroys the specified kernel QP.
3687 * @qp: The QP to destroy.
3689 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3691 static inline int ib_destroy_qp(struct ib_qp *qp)
3693 return ib_destroy_qp_user(qp, NULL);
3697 * ib_open_qp - Obtain a reference to an existing sharable QP.
3698 * @xrcd - XRC domain
3699 * @qp_open_attr: Attributes identifying the QP to open.
3701 * Returns a reference to a sharable QP.
3703 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3704 struct ib_qp_open_attr *qp_open_attr);
3707 * ib_close_qp - Release an external reference to a QP.
3708 * @qp: The QP handle to release
3710 * The opened QP handle is released by the caller. The underlying
3711 * shared QP is not destroyed until all internal references are released.
3713 int ib_close_qp(struct ib_qp *qp);
3716 * ib_post_send - Posts a list of work requests to the send queue of
3718 * @qp: The QP to post the work request on.
3719 * @send_wr: A list of work requests to post on the send queue.
3720 * @bad_send_wr: On an immediate failure, this parameter will reference
3721 * the work request that failed to be posted on the QP.
3723 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3724 * error is returned, the QP state shall not be affected,
3725 * ib_post_send() will return an immediate error after queueing any
3726 * earlier work requests in the list.
3728 static inline int ib_post_send(struct ib_qp *qp,
3729 const struct ib_send_wr *send_wr,
3730 const struct ib_send_wr **bad_send_wr)
3732 const struct ib_send_wr *dummy;
3734 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3738 * ib_post_recv - Posts a list of work requests to the receive queue of
3740 * @qp: The QP to post the work request on.
3741 * @recv_wr: A list of work requests to post on the receive queue.
3742 * @bad_recv_wr: On an immediate failure, this parameter will reference
3743 * the work request that failed to be posted on the QP.
3745 static inline int ib_post_recv(struct ib_qp *qp,
3746 const struct ib_recv_wr *recv_wr,
3747 const struct ib_recv_wr **bad_recv_wr)
3749 const struct ib_recv_wr *dummy;
3751 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3754 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3755 int nr_cqe, int comp_vector,
3756 enum ib_poll_context poll_ctx,
3757 const char *caller, struct ib_udata *udata);
3760 * ib_alloc_cq_user: Allocate kernel/user CQ
3761 * @dev: The IB device
3762 * @private: Private data attached to the CQE
3763 * @nr_cqe: Number of CQEs in the CQ
3764 * @comp_vector: Completion vector used for the IRQs
3765 * @poll_ctx: Context used for polling the CQ
3766 * @udata: Valid user data or NULL for kernel objects
3768 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3769 void *private, int nr_cqe,
3771 enum ib_poll_context poll_ctx,
3772 struct ib_udata *udata)
3774 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3775 KBUILD_MODNAME, udata);
3779 * ib_alloc_cq: Allocate kernel CQ
3780 * @dev: The IB device
3781 * @private: Private data attached to the CQE
3782 * @nr_cqe: Number of CQEs in the CQ
3783 * @comp_vector: Completion vector used for the IRQs
3784 * @poll_ctx: Context used for polling the CQ
3786 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3788 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3789 int nr_cqe, int comp_vector,
3790 enum ib_poll_context poll_ctx)
3792 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3796 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3797 int nr_cqe, enum ib_poll_context poll_ctx,
3798 const char *caller);
3801 * ib_alloc_cq_any: Allocate kernel CQ
3802 * @dev: The IB device
3803 * @private: Private data attached to the CQE
3804 * @nr_cqe: Number of CQEs in the CQ
3805 * @poll_ctx: Context used for polling the CQ
3807 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3808 void *private, int nr_cqe,
3809 enum ib_poll_context poll_ctx)
3811 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3816 * ib_free_cq_user - Free kernel/user CQ
3817 * @cq: The CQ to free
3818 * @udata: Valid user data or NULL for kernel objects
3820 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3823 * ib_free_cq - Free kernel CQ
3824 * @cq: The CQ to free
3826 * NOTE: for user cq use ib_free_cq_user with valid udata!
3828 static inline void ib_free_cq(struct ib_cq *cq)
3830 ib_free_cq_user(cq, NULL);
3833 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3836 * ib_create_cq - Creates a CQ on the specified device.
3837 * @device: The device on which to create the CQ.
3838 * @comp_handler: A user-specified callback that is invoked when a
3839 * completion event occurs on the CQ.
3840 * @event_handler: A user-specified callback that is invoked when an
3841 * asynchronous event not associated with a completion occurs on the CQ.
3842 * @cq_context: Context associated with the CQ returned to the user via
3843 * the associated completion and event handlers.
3844 * @cq_attr: The attributes the CQ should be created upon.
3846 * Users can examine the cq structure to determine the actual CQ size.
3848 struct ib_cq *__ib_create_cq(struct ib_device *device,
3849 ib_comp_handler comp_handler,
3850 void (*event_handler)(struct ib_event *, void *),
3852 const struct ib_cq_init_attr *cq_attr,
3853 const char *caller);
3854 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3855 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3858 * ib_resize_cq - Modifies the capacity of the CQ.
3859 * @cq: The CQ to resize.
3860 * @cqe: The minimum size of the CQ.
3862 * Users can examine the cq structure to determine the actual CQ size.
3864 int ib_resize_cq(struct ib_cq *cq, int cqe);
3867 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3868 * @cq: The CQ to modify.
3869 * @cq_count: number of CQEs that will trigger an event
3870 * @cq_period: max period of time in usec before triggering an event
3873 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3876 * ib_destroy_cq_user - Destroys the specified CQ.
3877 * @cq: The CQ to destroy.
3878 * @udata: Valid user data or NULL for kernel objects
3880 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3883 * ib_destroy_cq - Destroys the specified kernel CQ.
3884 * @cq: The CQ to destroy.
3886 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3888 static inline void ib_destroy_cq(struct ib_cq *cq)
3890 ib_destroy_cq_user(cq, NULL);
3894 * ib_poll_cq - poll a CQ for completion(s)
3895 * @cq:the CQ being polled
3896 * @num_entries:maximum number of completions to return
3897 * @wc:array of at least @num_entries &struct ib_wc where completions
3900 * Poll a CQ for (possibly multiple) completions. If the return value
3901 * is < 0, an error occurred. If the return value is >= 0, it is the
3902 * number of completions returned. If the return value is
3903 * non-negative and < num_entries, then the CQ was emptied.
3905 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3908 return cq->device->ops.poll_cq(cq, num_entries, wc);
3912 * ib_req_notify_cq - Request completion notification on a CQ.
3913 * @cq: The CQ to generate an event for.
3915 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3916 * to request an event on the next solicited event or next work
3917 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3918 * may also be |ed in to request a hint about missed events, as
3922 * < 0 means an error occurred while requesting notification
3923 * == 0 means notification was requested successfully, and if
3924 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3925 * were missed and it is safe to wait for another event. In
3926 * this case is it guaranteed that any work completions added
3927 * to the CQ since the last CQ poll will trigger a completion
3928 * notification event.
3929 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3930 * in. It means that the consumer must poll the CQ again to
3931 * make sure it is empty to avoid missing an event because of a
3932 * race between requesting notification and an entry being
3933 * added to the CQ. This return value means it is possible
3934 * (but not guaranteed) that a work completion has been added
3935 * to the CQ since the last poll without triggering a
3936 * completion notification event.
3938 static inline int ib_req_notify_cq(struct ib_cq *cq,
3939 enum ib_cq_notify_flags flags)
3941 return cq->device->ops.req_notify_cq(cq, flags);
3945 * ib_req_ncomp_notif - Request completion notification when there are
3946 * at least the specified number of unreaped completions on the CQ.
3947 * @cq: The CQ to generate an event for.
3948 * @wc_cnt: The number of unreaped completions that should be on the
3949 * CQ before an event is generated.
3951 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3953 return cq->device->ops.req_ncomp_notif ?
3954 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3959 * ib_dma_mapping_error - check a DMA addr for error
3960 * @dev: The device for which the dma_addr was created
3961 * @dma_addr: The DMA address to check
3963 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3965 return dma_mapping_error(dev->dma_device, dma_addr);
3969 * ib_dma_map_single - Map a kernel virtual address to DMA address
3970 * @dev: The device for which the dma_addr is to be created
3971 * @cpu_addr: The kernel virtual address
3972 * @size: The size of the region in bytes
3973 * @direction: The direction of the DMA
3975 static inline u64 ib_dma_map_single(struct ib_device *dev,
3976 void *cpu_addr, size_t size,
3977 enum dma_data_direction direction)
3979 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3983 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3984 * @dev: The device for which the DMA address was created
3985 * @addr: The DMA address
3986 * @size: The size of the region in bytes
3987 * @direction: The direction of the DMA
3989 static inline void ib_dma_unmap_single(struct ib_device *dev,
3990 u64 addr, size_t size,
3991 enum dma_data_direction direction)
3993 dma_unmap_single(dev->dma_device, addr, size, direction);
3997 * ib_dma_map_page - Map a physical page to DMA address
3998 * @dev: The device for which the dma_addr is to be created
3999 * @page: The page to be mapped
4000 * @offset: The offset within the page
4001 * @size: The size of the region in bytes
4002 * @direction: The direction of the DMA
4004 static inline u64 ib_dma_map_page(struct ib_device *dev,
4006 unsigned long offset,
4008 enum dma_data_direction direction)
4010 return dma_map_page(dev->dma_device, page, offset, size, direction);
4014 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4015 * @dev: The device for which the DMA address was created
4016 * @addr: The DMA address
4017 * @size: The size of the region in bytes
4018 * @direction: The direction of the DMA
4020 static inline void ib_dma_unmap_page(struct ib_device *dev,
4021 u64 addr, size_t size,
4022 enum dma_data_direction direction)
4024 dma_unmap_page(dev->dma_device, addr, size, direction);
4028 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4029 * @dev: The device for which the DMA addresses are to be created
4030 * @sg: The array of scatter/gather entries
4031 * @nents: The number of scatter/gather entries
4032 * @direction: The direction of the DMA
4034 static inline int ib_dma_map_sg(struct ib_device *dev,
4035 struct scatterlist *sg, int nents,
4036 enum dma_data_direction direction)
4038 return dma_map_sg(dev->dma_device, sg, nents, direction);
4042 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4043 * @dev: The device for which the DMA addresses were created
4044 * @sg: The array of scatter/gather entries
4045 * @nents: The number of scatter/gather entries
4046 * @direction: The direction of the DMA
4048 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4049 struct scatterlist *sg, int nents,
4050 enum dma_data_direction direction)
4052 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4055 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4056 struct scatterlist *sg, int nents,
4057 enum dma_data_direction direction,
4058 unsigned long dma_attrs)
4060 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4064 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4065 struct scatterlist *sg, int nents,
4066 enum dma_data_direction direction,
4067 unsigned long dma_attrs)
4069 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4073 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4074 * @dev: The device to query
4076 * The returned value represents a size in bytes.
4078 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4080 return dma_get_max_seg_size(dev->dma_device);
4084 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4085 * @dev: The device for which the DMA address was created
4086 * @addr: The DMA address
4087 * @size: The size of the region in bytes
4088 * @dir: The direction of the DMA
4090 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4093 enum dma_data_direction dir)
4095 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4099 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4100 * @dev: The device for which the DMA address was created
4101 * @addr: The DMA address
4102 * @size: The size of the region in bytes
4103 * @dir: The direction of the DMA
4105 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4108 enum dma_data_direction dir)
4110 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4114 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4115 * @dev: The device for which the DMA address is requested
4116 * @size: The size of the region to allocate in bytes
4117 * @dma_handle: A pointer for returning the DMA address of the region
4118 * @flag: memory allocator flags
4120 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4122 dma_addr_t *dma_handle,
4125 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4129 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4130 * @dev: The device for which the DMA addresses were allocated
4131 * @size: The size of the region
4132 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4133 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4135 static inline void ib_dma_free_coherent(struct ib_device *dev,
4136 size_t size, void *cpu_addr,
4137 dma_addr_t dma_handle)
4139 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4142 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4143 * space. This function should be called when 'current' is the owning MM.
4145 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4146 u64 virt_addr, int mr_access_flags);
4148 /* ib_advise_mr - give an advice about an address range in a memory region */
4149 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4150 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4152 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4153 * HCA translation table.
4154 * @mr: The memory region to deregister.
4155 * @udata: Valid user data or NULL for kernel object
4157 * This function can fail, if the memory region has memory windows bound to it.
4159 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4162 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4163 * HCA translation table.
4164 * @mr: The memory region to deregister.
4166 * This function can fail, if the memory region has memory windows bound to it.
4168 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4170 static inline int ib_dereg_mr(struct ib_mr *mr)
4172 return ib_dereg_mr_user(mr, NULL);
4175 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4176 u32 max_num_sg, struct ib_udata *udata);
4178 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4179 enum ib_mr_type mr_type, u32 max_num_sg)
4181 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4184 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4185 u32 max_num_data_sg,
4186 u32 max_num_meta_sg);
4189 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4191 * @mr - struct ib_mr pointer to be updated.
4192 * @newkey - new key to be used.
4194 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4196 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4197 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4201 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4202 * for calculating a new rkey for type 2 memory windows.
4203 * @rkey - the rkey to increment.
4205 static inline u32 ib_inc_rkey(u32 rkey)
4207 const u32 mask = 0x000000ff;
4208 return ((rkey + 1) & mask) | (rkey & ~mask);
4212 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4213 * @pd: The protection domain associated with the unmapped region.
4214 * @mr_access_flags: Specifies the memory access rights.
4215 * @fmr_attr: Attributes of the unmapped region.
4217 * A fast memory region must be mapped before it can be used as part of
4220 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4221 int mr_access_flags,
4222 struct ib_fmr_attr *fmr_attr);
4225 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4226 * @fmr: The fast memory region to associate with the pages.
4227 * @page_list: An array of physical pages to map to the fast memory region.
4228 * @list_len: The number of pages in page_list.
4229 * @iova: The I/O virtual address to use with the mapped region.
4231 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4232 u64 *page_list, int list_len,
4235 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4239 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4240 * @fmr_list: A linked list of fast memory regions to unmap.
4242 int ib_unmap_fmr(struct list_head *fmr_list);
4245 * ib_dealloc_fmr - Deallocates a fast memory region.
4246 * @fmr: The fast memory region to deallocate.
4248 int ib_dealloc_fmr(struct ib_fmr *fmr);
4251 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4252 * @qp: QP to attach to the multicast group. The QP must be type
4254 * @gid: Multicast group GID.
4255 * @lid: Multicast group LID in host byte order.
4257 * In order to send and receive multicast packets, subnet
4258 * administration must have created the multicast group and configured
4259 * the fabric appropriately. The port associated with the specified
4260 * QP must also be a member of the multicast group.
4262 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4265 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4266 * @qp: QP to detach from the multicast group.
4267 * @gid: Multicast group GID.
4268 * @lid: Multicast group LID in host byte order.
4270 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4273 * ib_alloc_xrcd - Allocates an XRC domain.
4274 * @device: The device on which to allocate the XRC domain.
4275 * @caller: Module name for kernel consumers
4277 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4278 #define ib_alloc_xrcd(device) \
4279 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4282 * ib_dealloc_xrcd - Deallocates an XRC domain.
4283 * @xrcd: The XRC domain to deallocate.
4284 * @udata: Valid user data or NULL for kernel object
4286 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4288 static inline int ib_check_mr_access(int flags)
4291 * Local write permission is required if remote write or
4292 * remote atomic permission is also requested.
4294 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4295 !(flags & IB_ACCESS_LOCAL_WRITE))
4298 if (flags & ~IB_ACCESS_SUPPORTED)
4304 static inline bool ib_access_writable(int access_flags)
4307 * We have writable memory backing the MR if any of the following
4308 * access flags are set. "Local write" and "remote write" obviously
4309 * require write access. "Remote atomic" can do things like fetch and
4310 * add, which will modify memory, and "MW bind" can change permissions
4311 * by binding a window.
4313 return access_flags &
4314 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4315 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4319 * ib_check_mr_status: lightweight check of MR status.
4320 * This routine may provide status checks on a selected
4321 * ib_mr. first use is for signature status check.
4323 * @mr: A memory region.
4324 * @check_mask: Bitmask of which checks to perform from
4325 * ib_mr_status_check enumeration.
4326 * @mr_status: The container of relevant status checks.
4327 * failed checks will be indicated in the status bitmask
4328 * and the relevant info shall be in the error item.
4330 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4331 struct ib_mr_status *mr_status);
4334 * ib_device_try_get: Hold a registration lock
4335 * device: The device to lock
4337 * A device under an active registration lock cannot become unregistered. It
4338 * is only possible to obtain a registration lock on a device that is fully
4339 * registered, otherwise this function returns false.
4341 * The registration lock is only necessary for actions which require the
4342 * device to still be registered. Uses that only require the device pointer to
4343 * be valid should use get_device(&ibdev->dev) to hold the memory.
4346 static inline bool ib_device_try_get(struct ib_device *dev)
4348 return refcount_inc_not_zero(&dev->refcount);
4351 void ib_device_put(struct ib_device *device);
4352 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4353 enum rdma_driver_id driver_id);
4354 struct ib_device *ib_device_get_by_name(const char *name,
4355 enum rdma_driver_id driver_id);
4356 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4357 u16 pkey, const union ib_gid *gid,
4358 const struct sockaddr *addr);
4359 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4361 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4363 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4364 struct ib_wq_init_attr *init_attr);
4365 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4366 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4368 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4369 struct ib_rwq_ind_table_init_attr*
4370 wq_ind_table_init_attr);
4371 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4373 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4374 unsigned int *sg_offset, unsigned int page_size);
4375 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4376 int data_sg_nents, unsigned int *data_sg_offset,
4377 struct scatterlist *meta_sg, int meta_sg_nents,
4378 unsigned int *meta_sg_offset, unsigned int page_size);
4381 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4382 unsigned int *sg_offset, unsigned int page_size)
4386 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4392 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4393 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4395 void ib_drain_rq(struct ib_qp *qp);
4396 void ib_drain_sq(struct ib_qp *qp);
4397 void ib_drain_qp(struct ib_qp *qp);
4399 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4401 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4403 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4404 return attr->roce.dmac;
4408 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4410 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4411 attr->ib.dlid = (u16)dlid;
4412 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4413 attr->opa.dlid = dlid;
4416 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4418 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4419 return attr->ib.dlid;
4420 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4421 return attr->opa.dlid;
4425 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4430 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4435 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4438 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4439 attr->ib.src_path_bits = src_path_bits;
4440 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4441 attr->opa.src_path_bits = src_path_bits;
4444 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4446 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4447 return attr->ib.src_path_bits;
4448 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4449 return attr->opa.src_path_bits;
4453 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4456 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4457 attr->opa.make_grd = make_grd;
4460 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4462 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4463 return attr->opa.make_grd;
4467 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4469 attr->port_num = port_num;
4472 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4474 return attr->port_num;
4477 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4480 attr->static_rate = static_rate;
4483 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4485 return attr->static_rate;
4488 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4489 enum ib_ah_flags flag)
4491 attr->ah_flags = flag;
4494 static inline enum ib_ah_flags
4495 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4497 return attr->ah_flags;
4500 static inline const struct ib_global_route
4501 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4506 /*To retrieve and modify the grh */
4507 static inline struct ib_global_route
4508 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4513 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4515 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4517 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4520 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4523 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4525 grh->dgid.global.subnet_prefix = prefix;
4528 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4531 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4533 grh->dgid.global.interface_id = if_id;
4536 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4537 union ib_gid *dgid, u32 flow_label,
4538 u8 sgid_index, u8 hop_limit,
4541 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4543 attr->ah_flags = IB_AH_GRH;
4546 grh->flow_label = flow_label;
4547 grh->sgid_index = sgid_index;
4548 grh->hop_limit = hop_limit;
4549 grh->traffic_class = traffic_class;
4550 grh->sgid_attr = NULL;
4553 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4554 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4555 u32 flow_label, u8 hop_limit, u8 traffic_class,
4556 const struct ib_gid_attr *sgid_attr);
4557 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4558 const struct rdma_ah_attr *src);
4559 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4560 const struct rdma_ah_attr *new);
4561 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4564 * rdma_ah_find_type - Return address handle type.
4566 * @dev: Device to be checked
4567 * @port_num: Port number
4569 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4572 if (rdma_protocol_roce(dev, port_num))
4573 return RDMA_AH_ATTR_TYPE_ROCE;
4574 if (rdma_protocol_ib(dev, port_num)) {
4575 if (rdma_cap_opa_ah(dev, port_num))
4576 return RDMA_AH_ATTR_TYPE_OPA;
4577 return RDMA_AH_ATTR_TYPE_IB;
4580 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4584 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4585 * In the current implementation the only way to get
4586 * get the 32bit lid is from other sources for OPA.
4587 * For IB, lids will always be 16bits so cast the
4588 * value accordingly.
4592 static inline u16 ib_lid_cpu16(u32 lid)
4594 WARN_ON_ONCE(lid & 0xFFFF0000);
4599 * ib_lid_be16 - Return lid in 16bit BE encoding.
4603 static inline __be16 ib_lid_be16(u32 lid)
4605 WARN_ON_ONCE(lid & 0xFFFF0000);
4606 return cpu_to_be16((u16)lid);
4610 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4612 * @device: the rdma device
4613 * @comp_vector: index of completion vector
4615 * Returns NULL on failure, otherwise a corresponding cpu map of the
4616 * completion vector (returns all-cpus map if the device driver doesn't
4617 * implement get_vector_affinity).
4619 static inline const struct cpumask *
4620 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4622 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4623 !device->ops.get_vector_affinity)
4626 return device->ops.get_vector_affinity(device, comp_vector);
4631 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4632 * and add their gids, as needed, to the relevant RoCE devices.
4634 * @device: the rdma device
4636 void rdma_roce_rescan_device(struct ib_device *ibdev);
4638 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4640 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4642 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4643 enum rdma_netdev_t type, const char *name,
4644 unsigned char name_assign_type,
4645 void (*setup)(struct net_device *));
4647 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4648 enum rdma_netdev_t type, const char *name,
4649 unsigned char name_assign_type,
4650 void (*setup)(struct net_device *),
4651 struct net_device *netdev);
4654 * rdma_set_device_sysfs_group - Set device attributes group to have
4655 * driver specific sysfs entries at
4656 * for infiniband class.
4658 * @device: device pointer for which attributes to be created
4659 * @group: Pointer to group which should be added when device
4660 * is registered with sysfs.
4661 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4662 * group per device to have sysfs attributes.
4664 * NOTE: New drivers should not make use of this API; instead new device
4665 * parameter should be exposed via netlink command. This API and mechanism
4666 * exist only for existing drivers.
4669 rdma_set_device_sysfs_group(struct ib_device *dev,
4670 const struct attribute_group *group)
4672 dev->groups[1] = group;
4676 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4678 * @device: device pointer for which ib_device pointer to retrieve
4680 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4683 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4685 struct ib_core_device *coredev =
4686 container_of(device, struct ib_core_device, dev);
4688 return coredev->owner;
4692 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4693 * ib_device holder structure from device pointer.
4695 * NOTE: New drivers should not make use of this API; This API is only for
4696 * existing drivers who have exposed sysfs entries using
4697 * rdma_set_device_sysfs_group().
4699 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4700 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4702 bool rdma_dev_access_netns(const struct ib_device *device,
4703 const struct net *net);
4704 #endif /* IB_VERBS_H */