1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
3 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
5 * Copyright (c) 2004, 2020 Intel Corporation. All rights reserved.
6 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
7 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
8 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
46 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
50 struct ib_usrq_object;
54 struct hw_stats_device_data;
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 const char *format, ...);
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...) \
83 dynamic_ibdev_dbg(__dev, format, ##args)
87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
92 static DEFINE_RATELIMIT_STATE(_rs, \
93 DEFAULT_RATELIMIT_INTERVAL, \
94 DEFAULT_RATELIMIT_BURST); \
95 if (__ratelimit(&_rs)) \
96 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
119 static DEFINE_RATELIMIT_STATE(_rs, \
120 DEFAULT_RATELIMIT_INTERVAL, \
121 DEFAULT_RATELIMIT_BURST); \
122 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
123 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
124 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
128 __printf(2, 3) __cold
130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
136 __be64 subnet_prefix;
141 extern union ib_gid zgid;
144 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
150 #define ROCE_V2_UDP_DPORT 4791
152 struct net_device __rcu *ndev;
153 struct ib_device *device;
155 enum ib_gid_type gid_type;
161 /* set the local administered indication */
162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
165 enum rdma_transport_type {
167 RDMA_TRANSPORT_IWARP,
168 RDMA_TRANSPORT_USNIC,
169 RDMA_TRANSPORT_USNIC_UDP,
170 RDMA_TRANSPORT_UNSPECIFIED,
173 enum rdma_protocol_type {
177 RDMA_PROTOCOL_USNIC_UDP
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
183 enum rdma_network_type {
185 RDMA_NETWORK_ROCE_V1,
190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
192 if (network_type == RDMA_NETWORK_IPV4 ||
193 network_type == RDMA_NETWORK_IPV6)
194 return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 else if (network_type == RDMA_NETWORK_ROCE_V1)
196 return IB_GID_TYPE_ROCE;
198 return IB_GID_TYPE_IB;
201 static inline enum rdma_network_type
202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
204 if (attr->gid_type == IB_GID_TYPE_IB)
205 return RDMA_NETWORK_IB;
207 if (attr->gid_type == IB_GID_TYPE_ROCE)
208 return RDMA_NETWORK_ROCE_V1;
210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 return RDMA_NETWORK_IPV4;
213 return RDMA_NETWORK_IPV6;
216 enum rdma_link_layer {
217 IB_LINK_LAYER_UNSPECIFIED,
218 IB_LINK_LAYER_INFINIBAND,
219 IB_LINK_LAYER_ETHERNET,
222 enum ib_device_cap_flags {
223 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
224 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
225 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
226 IB_DEVICE_RAW_MULTI = (1 << 3),
227 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
228 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
229 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
230 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
231 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
232 /* Not in use, former INIT_TYPE = (1 << 9),*/
233 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
234 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
235 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
236 IB_DEVICE_SRQ_RESIZE = (1 << 13),
237 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
240 * This device supports a per-device lkey or stag that can be
241 * used without performing a memory registration for the local
242 * memory. Note that ULPs should never check this flag, but
243 * instead of use the local_dma_lkey flag in the ib_pd structure,
244 * which will always contain a usable lkey.
246 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
247 /* Reserved, old SEND_W_INV = (1 << 16),*/
248 IB_DEVICE_MEM_WINDOW = (1 << 17),
250 * Devices should set IB_DEVICE_UD_IP_SUM if they support
251 * insertion of UDP and TCP checksum on outgoing UD IPoIB
252 * messages and can verify the validity of checksum for
253 * incoming messages. Setting this flag implies that the
254 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
256 IB_DEVICE_UD_IP_CSUM = (1 << 18),
257 IB_DEVICE_UD_TSO = (1 << 19),
258 IB_DEVICE_XRC = (1 << 20),
261 * This device supports the IB "base memory management extension",
262 * which includes support for fast registrations (IB_WR_REG_MR,
263 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
264 * also be set by any iWarp device which must support FRs to comply
265 * to the iWarp verbs spec. iWarp devices also support the
266 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
269 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
270 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
271 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
272 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
273 IB_DEVICE_RC_IP_CSUM = (1 << 25),
274 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
275 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
277 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
278 * support execution of WQEs that involve synchronization
279 * of I/O operations with single completion queue managed
282 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
283 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
284 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
285 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
286 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
287 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
288 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
289 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
290 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
291 /* The device supports padding incoming writes to cacheline. */
292 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
293 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
302 enum ib_odp_general_cap_bits {
303 IB_ODP_SUPPORT = 1 << 0,
304 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
307 enum ib_odp_transport_cap_bits {
308 IB_ODP_SUPPORT_SEND = 1 << 0,
309 IB_ODP_SUPPORT_RECV = 1 << 1,
310 IB_ODP_SUPPORT_WRITE = 1 << 2,
311 IB_ODP_SUPPORT_READ = 1 << 3,
312 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
313 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
317 uint64_t general_caps;
319 uint32_t rc_odp_caps;
320 uint32_t uc_odp_caps;
321 uint32_t ud_odp_caps;
322 uint32_t xrc_odp_caps;
323 } per_transport_caps;
327 /* Corresponding bit will be set if qp type from
328 * 'enum ib_qp_type' is supported, e.g.
329 * supported_qpts |= 1 << IB_QPT_UD
332 u32 max_rwq_indirection_tables;
333 u32 max_rwq_indirection_table_size;
336 enum ib_tm_cap_flags {
337 /* Support tag matching with rendezvous offload for RC transport */
338 IB_TM_CAP_RNDV_RC = 1 << 0,
342 /* Max size of RNDV header */
343 u32 max_rndv_hdr_size;
344 /* Max number of entries in tag matching list */
346 /* From enum ib_tm_cap_flags */
348 /* Max number of outstanding list operations */
350 /* Max number of SGE in tag matching entry */
354 struct ib_cq_init_attr {
360 enum ib_cq_attr_mask {
361 IB_CQ_MODERATE = 1 << 0,
365 u16 max_cq_moderation_count;
366 u16 max_cq_moderation_period;
369 struct ib_dm_mr_attr {
375 struct ib_dm_alloc_attr {
381 struct ib_device_attr {
383 __be64 sys_image_guid;
391 u64 device_cap_flags;
402 int max_qp_init_rd_atom;
403 int max_ee_init_rd_atom;
404 enum ib_atomic_cap atomic_cap;
405 enum ib_atomic_cap masked_atomic_cap;
412 int max_mcast_qp_attach;
413 int max_total_mcast_qp_attach;
418 unsigned int max_fast_reg_page_list_len;
419 unsigned int max_pi_fast_reg_page_list_len;
421 u8 local_ca_ack_delay;
424 struct ib_odp_caps odp_caps;
425 uint64_t timestamp_mask;
426 uint64_t hca_core_clock; /* in KHZ */
427 struct ib_rss_caps rss_caps;
429 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
430 struct ib_tm_caps tm_caps;
431 struct ib_cq_caps cq_caps;
433 /* Max entries for sgl for optimized performance per READ */
450 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
453 case IB_MTU_256: return 256;
454 case IB_MTU_512: return 512;
455 case IB_MTU_1024: return 1024;
456 case IB_MTU_2048: return 2048;
457 case IB_MTU_4096: return 4096;
462 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
466 else if (mtu >= 2048)
468 else if (mtu >= 1024)
476 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
484 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
488 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
491 return OPA_MTU_10240;
492 else if (mtu >= 8192)
495 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
504 IB_PORT_ACTIVE_DEFER = 5
507 enum ib_port_phys_state {
508 IB_PORT_PHYS_STATE_SLEEP = 1,
509 IB_PORT_PHYS_STATE_POLLING = 2,
510 IB_PORT_PHYS_STATE_DISABLED = 3,
511 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
512 IB_PORT_PHYS_STATE_LINK_UP = 5,
513 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
514 IB_PORT_PHYS_STATE_PHY_TEST = 7,
525 static inline int ib_width_enum_to_int(enum ib_port_width width)
528 case IB_WIDTH_1X: return 1;
529 case IB_WIDTH_2X: return 2;
530 case IB_WIDTH_4X: return 4;
531 case IB_WIDTH_8X: return 8;
532 case IB_WIDTH_12X: return 12;
549 * struct rdma_hw_stats
550 * @lock - Mutex to protect parallel write access to lifespan and values
551 * of counters, which are 64bits and not guaranteeed to be written
552 * atomicaly on 32bits systems.
553 * @timestamp - Used by the core code to track when the last update was
554 * @lifespan - Used by the core code to determine how old the counters
555 * should be before being updated again. Stored in jiffies, defaults
556 * to 10 milliseconds, drivers can override the default be specifying
557 * their own value during their allocation routine.
558 * @name - Array of pointers to static names used for the counters in
560 * @num_counters - How many hardware counters there are. If name is
561 * shorter than this number, a kernel oops will result. Driver authors
562 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
563 * in their code to prevent this.
564 * @value - Array of u64 counters that are accessed by the sysfs code and
565 * filled in by the drivers get_stats routine
567 struct rdma_hw_stats {
568 struct mutex lock; /* Protect lifespan and values[] */
569 unsigned long timestamp;
570 unsigned long lifespan;
571 const char * const *names;
576 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
578 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
580 * @names - Array of static const char *
581 * @num_counters - How many elements in array
582 * @lifespan - How many milliseconds between updates
584 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
585 const char * const *names, int num_counters,
586 unsigned long lifespan)
588 struct rdma_hw_stats *stats;
590 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
594 stats->names = names;
595 stats->num_counters = num_counters;
596 stats->lifespan = msecs_to_jiffies(lifespan);
602 /* Define bits for the various functionality this port needs to be supported by
605 /* Management 0x00000FFF */
606 #define RDMA_CORE_CAP_IB_MAD 0x00000001
607 #define RDMA_CORE_CAP_IB_SMI 0x00000002
608 #define RDMA_CORE_CAP_IB_CM 0x00000004
609 #define RDMA_CORE_CAP_IW_CM 0x00000008
610 #define RDMA_CORE_CAP_IB_SA 0x00000010
611 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
613 /* Address format 0x000FF000 */
614 #define RDMA_CORE_CAP_AF_IB 0x00001000
615 #define RDMA_CORE_CAP_ETH_AH 0x00002000
616 #define RDMA_CORE_CAP_OPA_AH 0x00004000
617 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
619 /* Protocol 0xFFF00000 */
620 #define RDMA_CORE_CAP_PROT_IB 0x00100000
621 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
622 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
623 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
624 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
625 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
627 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
628 | RDMA_CORE_CAP_PROT_ROCE \
629 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
631 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
632 | RDMA_CORE_CAP_IB_MAD \
633 | RDMA_CORE_CAP_IB_SMI \
634 | RDMA_CORE_CAP_IB_CM \
635 | RDMA_CORE_CAP_IB_SA \
636 | RDMA_CORE_CAP_AF_IB)
637 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
638 | RDMA_CORE_CAP_IB_MAD \
639 | RDMA_CORE_CAP_IB_CM \
640 | RDMA_CORE_CAP_AF_IB \
641 | RDMA_CORE_CAP_ETH_AH)
642 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
643 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
644 | RDMA_CORE_CAP_IB_MAD \
645 | RDMA_CORE_CAP_IB_CM \
646 | RDMA_CORE_CAP_AF_IB \
647 | RDMA_CORE_CAP_ETH_AH)
648 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
649 | RDMA_CORE_CAP_IW_CM)
650 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
651 | RDMA_CORE_CAP_OPA_MAD)
653 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
655 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
657 struct ib_port_attr {
659 enum ib_port_state state;
661 enum ib_mtu active_mtu;
664 unsigned int ip_gids:1;
665 /* This is the value from PortInfo CapabilityMask, defined by IBA */
684 enum ib_device_modify_flags {
685 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
686 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
689 #define IB_DEVICE_NODE_DESC_MAX 64
691 struct ib_device_modify {
693 char node_desc[IB_DEVICE_NODE_DESC_MAX];
696 enum ib_port_modify_flags {
697 IB_PORT_SHUTDOWN = 1,
698 IB_PORT_INIT_TYPE = (1<<2),
699 IB_PORT_RESET_QKEY_CNTR = (1<<3),
700 IB_PORT_OPA_MASK_CHG = (1<<4)
703 struct ib_port_modify {
704 u32 set_port_cap_mask;
705 u32 clr_port_cap_mask;
713 IB_EVENT_QP_ACCESS_ERR,
717 IB_EVENT_PATH_MIG_ERR,
718 IB_EVENT_DEVICE_FATAL,
719 IB_EVENT_PORT_ACTIVE,
722 IB_EVENT_PKEY_CHANGE,
725 IB_EVENT_SRQ_LIMIT_REACHED,
726 IB_EVENT_QP_LAST_WQE_REACHED,
727 IB_EVENT_CLIENT_REREGISTER,
732 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
735 struct ib_device *device;
743 enum ib_event_type event;
746 struct ib_event_handler {
747 struct ib_device *device;
748 void (*handler)(struct ib_event_handler *, struct ib_event *);
749 struct list_head list;
752 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
754 (_ptr)->device = _device; \
755 (_ptr)->handler = _handler; \
756 INIT_LIST_HEAD(&(_ptr)->list); \
759 struct ib_global_route {
760 const struct ib_gid_attr *sgid_attr;
769 __be32 version_tclass_flow;
777 union rdma_network_hdr {
780 /* The IB spec states that if it's IPv4, the header
781 * is located in the last 20 bytes of the header.
784 struct iphdr roce4grh;
788 #define IB_QPN_MASK 0xFFFFFF
791 IB_MULTICAST_QPN = 0xffffff
794 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
795 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
802 IB_RATE_PORT_CURRENT = 0,
803 IB_RATE_2_5_GBPS = 2,
811 IB_RATE_120_GBPS = 10,
812 IB_RATE_14_GBPS = 11,
813 IB_RATE_56_GBPS = 12,
814 IB_RATE_112_GBPS = 13,
815 IB_RATE_168_GBPS = 14,
816 IB_RATE_25_GBPS = 15,
817 IB_RATE_100_GBPS = 16,
818 IB_RATE_200_GBPS = 17,
819 IB_RATE_300_GBPS = 18,
820 IB_RATE_28_GBPS = 19,
821 IB_RATE_50_GBPS = 20,
822 IB_RATE_400_GBPS = 21,
823 IB_RATE_600_GBPS = 22,
827 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
828 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
829 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
830 * @rate: rate to convert.
832 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
835 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
836 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
837 * @rate: rate to convert.
839 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
843 * enum ib_mr_type - memory region type
844 * @IB_MR_TYPE_MEM_REG: memory region that is used for
845 * normal registration
846 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
847 * register any arbitrary sg lists (without
848 * the normal mr constraints - see
850 * @IB_MR_TYPE_DM: memory region that is used for device
851 * memory registration
852 * @IB_MR_TYPE_USER: memory region that is used for the user-space
854 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
855 * without address translations (VA=PA)
856 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
857 * data integrity operations
865 IB_MR_TYPE_INTEGRITY,
868 enum ib_mr_status_check {
869 IB_MR_CHECK_SIG_STATUS = 1,
873 * struct ib_mr_status - Memory region status container
875 * @fail_status: Bitmask of MR checks status. For each
876 * failed check a corresponding status bit is set.
877 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
880 struct ib_mr_status {
882 struct ib_sig_err sig_err;
886 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
888 * @mult: multiple to convert.
890 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
892 struct rdma_ah_init_attr {
893 struct rdma_ah_attr *ah_attr;
895 struct net_device *xmit_slave;
898 enum rdma_ah_attr_type {
899 RDMA_AH_ATTR_TYPE_UNDEFINED,
900 RDMA_AH_ATTR_TYPE_IB,
901 RDMA_AH_ATTR_TYPE_ROCE,
902 RDMA_AH_ATTR_TYPE_OPA,
910 struct roce_ah_attr {
920 struct rdma_ah_attr {
921 struct ib_global_route grh;
926 enum rdma_ah_attr_type type;
928 struct ib_ah_attr ib;
929 struct roce_ah_attr roce;
930 struct opa_ah_attr opa;
938 IB_WC_LOC_EEC_OP_ERR,
943 IB_WC_LOC_ACCESS_ERR,
944 IB_WC_REM_INV_REQ_ERR,
945 IB_WC_REM_ACCESS_ERR,
948 IB_WC_RNR_RETRY_EXC_ERR,
949 IB_WC_LOC_RDD_VIOL_ERR,
950 IB_WC_REM_INV_RD_REQ_ERR,
953 IB_WC_INV_EEC_STATE_ERR,
955 IB_WC_RESP_TIMEOUT_ERR,
959 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
962 IB_WC_SEND = IB_UVERBS_WC_SEND,
963 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
964 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
965 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
966 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
967 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
968 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
969 IB_WC_LSO = IB_UVERBS_WC_TSO,
971 IB_WC_MASKED_COMP_SWAP,
972 IB_WC_MASKED_FETCH_ADD,
974 * Set value of IB_WC_RECV so consumers can test if a completion is a
975 * receive by testing (opcode & IB_WC_RECV).
978 IB_WC_RECV_RDMA_WITH_IMM
983 IB_WC_WITH_IMM = (1<<1),
984 IB_WC_WITH_INVALIDATE = (1<<2),
985 IB_WC_IP_CSUM_OK = (1<<3),
986 IB_WC_WITH_SMAC = (1<<4),
987 IB_WC_WITH_VLAN = (1<<5),
988 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
994 struct ib_cqe *wr_cqe;
996 enum ib_wc_status status;
997 enum ib_wc_opcode opcode;
1003 u32 invalidate_rkey;
1011 u32 port_num; /* valid only for DR SMPs on switches */
1014 u8 network_hdr_type;
1017 enum ib_cq_notify_flags {
1018 IB_CQ_SOLICITED = 1 << 0,
1019 IB_CQ_NEXT_COMP = 1 << 1,
1020 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1021 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1025 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1026 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1027 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1030 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1032 return srq_type == IB_SRQT_XRC ||
1033 srq_type == IB_SRQT_TM;
1036 enum ib_srq_attr_mask {
1037 IB_SRQ_MAX_WR = 1 << 0,
1038 IB_SRQ_LIMIT = 1 << 1,
1041 struct ib_srq_attr {
1047 struct ib_srq_init_attr {
1048 void (*event_handler)(struct ib_event *, void *);
1050 struct ib_srq_attr attr;
1051 enum ib_srq_type srq_type;
1057 struct ib_xrcd *xrcd;
1072 u32 max_inline_data;
1075 * Maximum number of rdma_rw_ctx structures in flight at a time.
1076 * ib_create_qp() will calculate the right amount of neededed WRs
1077 * and MRs based on this.
1089 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1090 * here (and in that order) since the MAD layer uses them as
1091 * indices into a 2-entry table.
1096 IB_QPT_RC = IB_UVERBS_QPT_RC,
1097 IB_QPT_UC = IB_UVERBS_QPT_UC,
1098 IB_QPT_UD = IB_UVERBS_QPT_UD,
1100 IB_QPT_RAW_ETHERTYPE,
1101 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1102 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1103 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1105 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1106 /* Reserve a range for qp types internal to the low level driver.
1107 * These qp types will not be visible at the IB core layer, so the
1108 * IB_QPT_MAX usages should not be affected in the core layer
1110 IB_QPT_RESERVED1 = 0x1000,
1122 enum ib_qp_create_flags {
1123 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1124 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1125 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1126 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1127 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1128 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1129 IB_QP_CREATE_NETIF_QP = 1 << 5,
1130 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1131 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1132 IB_QP_CREATE_SCATTER_FCS =
1133 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1134 IB_QP_CREATE_CVLAN_STRIPPING =
1135 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1136 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1137 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1138 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1139 /* reserve bits 26-31 for low level drivers' internal use */
1140 IB_QP_CREATE_RESERVED_START = 1 << 26,
1141 IB_QP_CREATE_RESERVED_END = 1 << 31,
1145 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1146 * callback to destroy the passed in QP.
1149 struct ib_qp_init_attr {
1150 /* Consumer's event_handler callback must not block */
1151 void (*event_handler)(struct ib_event *, void *);
1154 struct ib_cq *send_cq;
1155 struct ib_cq *recv_cq;
1157 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1158 struct ib_qp_cap cap;
1159 enum ib_sig_type sq_sig_type;
1160 enum ib_qp_type qp_type;
1164 * Only needed for special QP types, or when using the RW API.
1167 struct ib_rwq_ind_table *rwq_ind_tbl;
1171 struct ib_qp_open_attr {
1172 void (*event_handler)(struct ib_event *, void *);
1175 enum ib_qp_type qp_type;
1178 enum ib_rnr_timeout {
1179 IB_RNR_TIMER_655_36 = 0,
1180 IB_RNR_TIMER_000_01 = 1,
1181 IB_RNR_TIMER_000_02 = 2,
1182 IB_RNR_TIMER_000_03 = 3,
1183 IB_RNR_TIMER_000_04 = 4,
1184 IB_RNR_TIMER_000_06 = 5,
1185 IB_RNR_TIMER_000_08 = 6,
1186 IB_RNR_TIMER_000_12 = 7,
1187 IB_RNR_TIMER_000_16 = 8,
1188 IB_RNR_TIMER_000_24 = 9,
1189 IB_RNR_TIMER_000_32 = 10,
1190 IB_RNR_TIMER_000_48 = 11,
1191 IB_RNR_TIMER_000_64 = 12,
1192 IB_RNR_TIMER_000_96 = 13,
1193 IB_RNR_TIMER_001_28 = 14,
1194 IB_RNR_TIMER_001_92 = 15,
1195 IB_RNR_TIMER_002_56 = 16,
1196 IB_RNR_TIMER_003_84 = 17,
1197 IB_RNR_TIMER_005_12 = 18,
1198 IB_RNR_TIMER_007_68 = 19,
1199 IB_RNR_TIMER_010_24 = 20,
1200 IB_RNR_TIMER_015_36 = 21,
1201 IB_RNR_TIMER_020_48 = 22,
1202 IB_RNR_TIMER_030_72 = 23,
1203 IB_RNR_TIMER_040_96 = 24,
1204 IB_RNR_TIMER_061_44 = 25,
1205 IB_RNR_TIMER_081_92 = 26,
1206 IB_RNR_TIMER_122_88 = 27,
1207 IB_RNR_TIMER_163_84 = 28,
1208 IB_RNR_TIMER_245_76 = 29,
1209 IB_RNR_TIMER_327_68 = 30,
1210 IB_RNR_TIMER_491_52 = 31
1213 enum ib_qp_attr_mask {
1215 IB_QP_CUR_STATE = (1<<1),
1216 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1217 IB_QP_ACCESS_FLAGS = (1<<3),
1218 IB_QP_PKEY_INDEX = (1<<4),
1219 IB_QP_PORT = (1<<5),
1220 IB_QP_QKEY = (1<<6),
1222 IB_QP_PATH_MTU = (1<<8),
1223 IB_QP_TIMEOUT = (1<<9),
1224 IB_QP_RETRY_CNT = (1<<10),
1225 IB_QP_RNR_RETRY = (1<<11),
1226 IB_QP_RQ_PSN = (1<<12),
1227 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1228 IB_QP_ALT_PATH = (1<<14),
1229 IB_QP_MIN_RNR_TIMER = (1<<15),
1230 IB_QP_SQ_PSN = (1<<16),
1231 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1232 IB_QP_PATH_MIG_STATE = (1<<18),
1233 IB_QP_CAP = (1<<19),
1234 IB_QP_DEST_QPN = (1<<20),
1235 IB_QP_RESERVED1 = (1<<21),
1236 IB_QP_RESERVED2 = (1<<22),
1237 IB_QP_RESERVED3 = (1<<23),
1238 IB_QP_RESERVED4 = (1<<24),
1239 IB_QP_RATE_LIMIT = (1<<25),
1241 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1266 enum ib_qp_state qp_state;
1267 enum ib_qp_state cur_qp_state;
1268 enum ib_mtu path_mtu;
1269 enum ib_mig_state path_mig_state;
1274 int qp_access_flags;
1275 struct ib_qp_cap cap;
1276 struct rdma_ah_attr ah_attr;
1277 struct rdma_ah_attr alt_ah_attr;
1280 u8 en_sqd_async_notify;
1283 u8 max_dest_rd_atomic;
1292 struct net_device *xmit_slave;
1296 /* These are shared with userspace */
1297 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1298 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1299 IB_WR_SEND = IB_UVERBS_WR_SEND,
1300 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1301 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1302 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1303 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1304 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1305 IB_WR_LSO = IB_UVERBS_WR_TSO,
1306 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1307 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1308 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1309 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1310 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1311 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1312 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1314 /* These are kernel only and can not be issued by userspace */
1315 IB_WR_REG_MR = 0x20,
1316 IB_WR_REG_MR_INTEGRITY,
1318 /* reserve values for low level drivers' internal use.
1319 * These values will not be used at all in the ib core layer.
1321 IB_WR_RESERVED1 = 0xf0,
1333 enum ib_send_flags {
1335 IB_SEND_SIGNALED = (1<<1),
1336 IB_SEND_SOLICITED = (1<<2),
1337 IB_SEND_INLINE = (1<<3),
1338 IB_SEND_IP_CSUM = (1<<4),
1340 /* reserve bits 26-31 for low level drivers' internal use */
1341 IB_SEND_RESERVED_START = (1 << 26),
1342 IB_SEND_RESERVED_END = (1 << 31),
1352 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1356 struct ib_send_wr *next;
1359 struct ib_cqe *wr_cqe;
1361 struct ib_sge *sg_list;
1363 enum ib_wr_opcode opcode;
1367 u32 invalidate_rkey;
1372 struct ib_send_wr wr;
1377 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1379 return container_of(wr, struct ib_rdma_wr, wr);
1382 struct ib_atomic_wr {
1383 struct ib_send_wr wr;
1387 u64 compare_add_mask;
1392 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1394 return container_of(wr, struct ib_atomic_wr, wr);
1398 struct ib_send_wr wr;
1405 u16 pkey_index; /* valid for GSI only */
1406 u32 port_num; /* valid for DR SMPs on switch only */
1409 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1411 return container_of(wr, struct ib_ud_wr, wr);
1415 struct ib_send_wr wr;
1421 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1423 return container_of(wr, struct ib_reg_wr, wr);
1427 struct ib_recv_wr *next;
1430 struct ib_cqe *wr_cqe;
1432 struct ib_sge *sg_list;
1436 enum ib_access_flags {
1437 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1438 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1439 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1440 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1441 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1442 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1443 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1444 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1445 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1447 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1448 IB_ACCESS_SUPPORTED =
1449 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1453 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1454 * are hidden here instead of a uapi header!
1456 enum ib_mr_rereg_flags {
1457 IB_MR_REREG_TRANS = 1,
1458 IB_MR_REREG_PD = (1<<1),
1459 IB_MR_REREG_ACCESS = (1<<2),
1460 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1465 enum rdma_remove_reason {
1467 * Userspace requested uobject deletion or initial try
1468 * to remove uobject via cleanup. Call could fail
1470 RDMA_REMOVE_DESTROY,
1471 /* Context deletion. This call should delete the actual object itself */
1473 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1474 RDMA_REMOVE_DRIVER_REMOVE,
1475 /* uobj is being cleaned-up before being committed */
1477 /* The driver failed to destroy the uobject and is being disconnected */
1478 RDMA_REMOVE_DRIVER_FAILURE,
1481 struct ib_rdmacg_object {
1482 #ifdef CONFIG_CGROUP_RDMA
1483 struct rdma_cgroup *cg; /* owner rdma cgroup */
1487 struct ib_ucontext {
1488 struct ib_device *device;
1489 struct ib_uverbs_file *ufile;
1491 struct ib_rdmacg_object cg_obj;
1493 * Implementation details of the RDMA core, don't use in drivers:
1495 struct rdma_restrack_entry res;
1496 struct xarray mmap_xa;
1500 u64 user_handle; /* handle given to us by userspace */
1501 /* ufile & ucontext owning this object */
1502 struct ib_uverbs_file *ufile;
1503 /* FIXME, save memory: ufile->context == context */
1504 struct ib_ucontext *context; /* associated user context */
1505 void *object; /* containing object */
1506 struct list_head list; /* link to context's list */
1507 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1508 int id; /* index into kernel idr */
1510 atomic_t usecnt; /* protects exclusive access */
1511 struct rcu_head rcu; /* kfree_rcu() overhead */
1513 const struct uverbs_api_object *uapi_object;
1517 const void __user *inbuf;
1518 void __user *outbuf;
1526 struct ib_device *device;
1527 struct ib_uobject *uobject;
1528 atomic_t usecnt; /* count all resources */
1530 u32 unsafe_global_rkey;
1533 * Implementation details of the RDMA core, don't use in drivers:
1535 struct ib_mr *__internal_mr;
1536 struct rdma_restrack_entry res;
1540 struct ib_device *device;
1541 atomic_t usecnt; /* count all exposed resources */
1542 struct inode *inode;
1543 struct rw_semaphore tgt_qps_rwsem;
1544 struct xarray tgt_qps;
1548 struct ib_device *device;
1550 struct ib_uobject *uobject;
1551 const struct ib_gid_attr *sgid_attr;
1552 enum rdma_ah_attr_type type;
1555 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1557 enum ib_poll_context {
1558 IB_POLL_SOFTIRQ, /* poll from softirq context */
1559 IB_POLL_WORKQUEUE, /* poll from workqueue */
1560 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1561 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1563 IB_POLL_DIRECT, /* caller context, no hw completions */
1567 struct ib_device *device;
1568 struct ib_ucq_object *uobject;
1569 ib_comp_handler comp_handler;
1570 void (*event_handler)(struct ib_event *, void *);
1573 unsigned int cqe_used;
1574 atomic_t usecnt; /* count number of work queues */
1575 enum ib_poll_context poll_ctx;
1577 struct list_head pool_entry;
1579 struct irq_poll iop;
1580 struct work_struct work;
1582 struct workqueue_struct *comp_wq;
1585 /* updated only by trace points */
1589 unsigned int comp_vector;
1592 * Implementation details of the RDMA core, don't use in drivers:
1594 struct rdma_restrack_entry res;
1598 struct ib_device *device;
1600 struct ib_usrq_object *uobject;
1601 void (*event_handler)(struct ib_event *, void *);
1603 enum ib_srq_type srq_type;
1610 struct ib_xrcd *xrcd;
1617 * Implementation details of the RDMA core, don't use in drivers:
1619 struct rdma_restrack_entry res;
1622 enum ib_raw_packet_caps {
1623 /* Strip cvlan from incoming packet and report it in the matching work
1624 * completion is supported.
1626 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1627 /* Scatter FCS field of an incoming packet to host memory is supported.
1629 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1630 /* Checksum offloads are supported (for both send and receive). */
1631 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1632 /* When a packet is received for an RQ with no receive WQEs, the
1633 * packet processing is delayed.
1635 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1639 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1649 struct ib_device *device;
1650 struct ib_uwq_object *uobject;
1652 void (*event_handler)(struct ib_event *, void *);
1656 enum ib_wq_state state;
1657 enum ib_wq_type wq_type;
1662 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1663 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1664 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1665 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1666 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1669 struct ib_wq_init_attr {
1671 enum ib_wq_type wq_type;
1675 void (*event_handler)(struct ib_event *, void *);
1676 u32 create_flags; /* Use enum ib_wq_flags */
1679 enum ib_wq_attr_mask {
1680 IB_WQ_STATE = 1 << 0,
1681 IB_WQ_CUR_STATE = 1 << 1,
1682 IB_WQ_FLAGS = 1 << 2,
1686 enum ib_wq_state wq_state;
1687 enum ib_wq_state curr_wq_state;
1688 u32 flags; /* Use enum ib_wq_flags */
1689 u32 flags_mask; /* Use enum ib_wq_flags */
1692 struct ib_rwq_ind_table {
1693 struct ib_device *device;
1694 struct ib_uobject *uobject;
1697 u32 log_ind_tbl_size;
1698 struct ib_wq **ind_tbl;
1701 struct ib_rwq_ind_table_init_attr {
1702 u32 log_ind_tbl_size;
1703 /* Each entry is a pointer to Receive Work Queue */
1704 struct ib_wq **ind_tbl;
1707 enum port_pkey_state {
1708 IB_PORT_PKEY_NOT_VALID = 0,
1709 IB_PORT_PKEY_VALID = 1,
1710 IB_PORT_PKEY_LISTED = 2,
1713 struct ib_qp_security;
1715 struct ib_port_pkey {
1716 enum port_pkey_state state;
1719 struct list_head qp_list;
1720 struct list_head to_error_list;
1721 struct ib_qp_security *sec;
1724 struct ib_ports_pkeys {
1725 struct ib_port_pkey main;
1726 struct ib_port_pkey alt;
1729 struct ib_qp_security {
1731 struct ib_device *dev;
1732 /* Hold this mutex when changing port and pkey settings. */
1734 struct ib_ports_pkeys *ports_pkeys;
1735 /* A list of all open shared QP handles. Required to enforce security
1736 * properly for all users of a shared QP.
1738 struct list_head shared_qp_list;
1741 atomic_t error_list_count;
1742 struct completion error_complete;
1743 int error_comps_pending;
1747 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1748 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1751 struct ib_device *device;
1753 struct ib_cq *send_cq;
1754 struct ib_cq *recv_cq;
1757 struct list_head rdma_mrs;
1758 struct list_head sig_mrs;
1760 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1761 struct list_head xrcd_list;
1763 /* count times opened, mcast attaches, flow attaches */
1765 struct list_head open_list;
1766 struct ib_qp *real_qp;
1767 struct ib_uqp_object *uobject;
1768 void (*event_handler)(struct ib_event *, void *);
1770 /* sgid_attrs associated with the AV's */
1771 const struct ib_gid_attr *av_sgid_attr;
1772 const struct ib_gid_attr *alt_path_sgid_attr;
1776 enum ib_qp_type qp_type;
1777 struct ib_rwq_ind_table *rwq_ind_tbl;
1778 struct ib_qp_security *qp_sec;
1783 * Implementation details of the RDMA core, don't use in drivers:
1785 struct rdma_restrack_entry res;
1787 /* The counter the qp is bind to */
1788 struct rdma_counter *counter;
1792 struct ib_device *device;
1795 struct ib_uobject *uobject;
1800 struct ib_device *device;
1806 unsigned int page_size;
1807 enum ib_mr_type type;
1810 struct ib_uobject *uobject; /* user */
1811 struct list_head qp_entry; /* FR */
1815 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1817 * Implementation details of the RDMA core, don't use in drivers:
1819 struct rdma_restrack_entry res;
1823 struct ib_device *device;
1825 struct ib_uobject *uobject;
1827 enum ib_mw_type type;
1830 /* Supported steering options */
1831 enum ib_flow_attr_type {
1832 /* steering according to rule specifications */
1833 IB_FLOW_ATTR_NORMAL = 0x0,
1834 /* default unicast and multicast rule -
1835 * receive all Eth traffic which isn't steered to any QP
1837 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1838 /* default multicast rule -
1839 * receive all Eth multicast traffic which isn't steered to any QP
1841 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1842 /* sniffer rule - receive all port traffic */
1843 IB_FLOW_ATTR_SNIFFER = 0x3
1846 /* Supported steering header types */
1847 enum ib_flow_spec_type {
1849 IB_FLOW_SPEC_ETH = 0x20,
1850 IB_FLOW_SPEC_IB = 0x22,
1852 IB_FLOW_SPEC_IPV4 = 0x30,
1853 IB_FLOW_SPEC_IPV6 = 0x31,
1854 IB_FLOW_SPEC_ESP = 0x34,
1856 IB_FLOW_SPEC_TCP = 0x40,
1857 IB_FLOW_SPEC_UDP = 0x41,
1858 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1859 IB_FLOW_SPEC_GRE = 0x51,
1860 IB_FLOW_SPEC_MPLS = 0x60,
1861 IB_FLOW_SPEC_INNER = 0x100,
1863 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1864 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1865 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1866 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1868 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1869 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1871 enum ib_flow_flags {
1872 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1873 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1874 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1877 struct ib_flow_eth_filter {
1886 struct ib_flow_spec_eth {
1889 struct ib_flow_eth_filter val;
1890 struct ib_flow_eth_filter mask;
1893 struct ib_flow_ib_filter {
1900 struct ib_flow_spec_ib {
1903 struct ib_flow_ib_filter val;
1904 struct ib_flow_ib_filter mask;
1907 /* IPv4 header flags */
1908 enum ib_ipv4_flags {
1909 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1910 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1911 last have this flag set */
1914 struct ib_flow_ipv4_filter {
1925 struct ib_flow_spec_ipv4 {
1928 struct ib_flow_ipv4_filter val;
1929 struct ib_flow_ipv4_filter mask;
1932 struct ib_flow_ipv6_filter {
1943 struct ib_flow_spec_ipv6 {
1946 struct ib_flow_ipv6_filter val;
1947 struct ib_flow_ipv6_filter mask;
1950 struct ib_flow_tcp_udp_filter {
1957 struct ib_flow_spec_tcp_udp {
1960 struct ib_flow_tcp_udp_filter val;
1961 struct ib_flow_tcp_udp_filter mask;
1964 struct ib_flow_tunnel_filter {
1969 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1970 * the tunnel_id from val has the vni value
1972 struct ib_flow_spec_tunnel {
1975 struct ib_flow_tunnel_filter val;
1976 struct ib_flow_tunnel_filter mask;
1979 struct ib_flow_esp_filter {
1986 struct ib_flow_spec_esp {
1989 struct ib_flow_esp_filter val;
1990 struct ib_flow_esp_filter mask;
1993 struct ib_flow_gre_filter {
1994 __be16 c_ks_res0_ver;
2001 struct ib_flow_spec_gre {
2004 struct ib_flow_gre_filter val;
2005 struct ib_flow_gre_filter mask;
2008 struct ib_flow_mpls_filter {
2014 struct ib_flow_spec_mpls {
2017 struct ib_flow_mpls_filter val;
2018 struct ib_flow_mpls_filter mask;
2021 struct ib_flow_spec_action_tag {
2022 enum ib_flow_spec_type type;
2027 struct ib_flow_spec_action_drop {
2028 enum ib_flow_spec_type type;
2032 struct ib_flow_spec_action_handle {
2033 enum ib_flow_spec_type type;
2035 struct ib_flow_action *act;
2038 enum ib_counters_description {
2043 struct ib_flow_spec_action_count {
2044 enum ib_flow_spec_type type;
2046 struct ib_counters *counters;
2049 union ib_flow_spec {
2054 struct ib_flow_spec_eth eth;
2055 struct ib_flow_spec_ib ib;
2056 struct ib_flow_spec_ipv4 ipv4;
2057 struct ib_flow_spec_tcp_udp tcp_udp;
2058 struct ib_flow_spec_ipv6 ipv6;
2059 struct ib_flow_spec_tunnel tunnel;
2060 struct ib_flow_spec_esp esp;
2061 struct ib_flow_spec_gre gre;
2062 struct ib_flow_spec_mpls mpls;
2063 struct ib_flow_spec_action_tag flow_tag;
2064 struct ib_flow_spec_action_drop drop;
2065 struct ib_flow_spec_action_handle action;
2066 struct ib_flow_spec_action_count flow_count;
2069 struct ib_flow_attr {
2070 enum ib_flow_attr_type type;
2076 union ib_flow_spec flows[];
2081 struct ib_device *device;
2082 struct ib_uobject *uobject;
2085 enum ib_flow_action_type {
2086 IB_FLOW_ACTION_UNSPECIFIED,
2087 IB_FLOW_ACTION_ESP = 1,
2090 struct ib_flow_action_attrs_esp_keymats {
2091 enum ib_uverbs_flow_action_esp_keymat protocol;
2093 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2097 struct ib_flow_action_attrs_esp_replays {
2098 enum ib_uverbs_flow_action_esp_replay protocol;
2100 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2104 enum ib_flow_action_attrs_esp_flags {
2105 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2106 * This is done in order to share the same flags between user-space and
2107 * kernel and spare an unnecessary translation.
2111 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2112 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2115 struct ib_flow_spec_list {
2116 struct ib_flow_spec_list *next;
2117 union ib_flow_spec spec;
2120 struct ib_flow_action_attrs_esp {
2121 struct ib_flow_action_attrs_esp_keymats *keymat;
2122 struct ib_flow_action_attrs_esp_replays *replay;
2123 struct ib_flow_spec_list *encap;
2124 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2125 * Value of 0 is a valid value.
2131 /* Use enum ib_flow_action_attrs_esp_flags */
2133 u64 hard_limit_pkts;
2136 struct ib_flow_action {
2137 struct ib_device *device;
2138 struct ib_uobject *uobject;
2139 enum ib_flow_action_type type;
2145 enum ib_process_mad_flags {
2146 IB_MAD_IGNORE_MKEY = 1,
2147 IB_MAD_IGNORE_BKEY = 2,
2148 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2151 enum ib_mad_result {
2152 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2153 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2154 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2155 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2158 struct ib_port_cache {
2160 struct ib_pkey_cache *pkey;
2161 struct ib_gid_table *gid;
2163 enum ib_port_state port_state;
2166 struct ib_port_immutable {
2173 struct ib_port_data {
2174 struct ib_device *ib_dev;
2176 struct ib_port_immutable immutable;
2178 spinlock_t pkey_list_lock;
2180 spinlock_t netdev_lock;
2182 struct list_head pkey_list;
2184 struct ib_port_cache cache;
2186 struct net_device __rcu *netdev;
2187 struct hlist_node ndev_hash_link;
2188 struct rdma_port_counter port_counter;
2189 struct ib_port *sysfs;
2192 /* rdma netdev type - specifies protocol type */
2193 enum rdma_netdev_t {
2194 RDMA_NETDEV_OPA_VNIC,
2199 * struct rdma_netdev - rdma netdev
2200 * For cases where netstack interfacing is required.
2202 struct rdma_netdev {
2204 struct ib_device *hca;
2209 * cleanup function must be specified.
2210 * FIXME: This is only used for OPA_VNIC and that usage should be
2213 void (*free_rdma_netdev)(struct net_device *netdev);
2215 /* control functions */
2216 void (*set_id)(struct net_device *netdev, int id);
2218 int (*send)(struct net_device *dev, struct sk_buff *skb,
2219 struct ib_ah *address, u32 dqpn);
2221 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2222 union ib_gid *gid, u16 mlid,
2223 int set_qkey, u32 qkey);
2224 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2225 union ib_gid *gid, u16 mlid);
2227 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2230 struct rdma_netdev_alloc_params {
2236 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2237 struct net_device *netdev, void *param);
2240 struct ib_odp_counters {
2242 atomic64_t invalidations;
2243 atomic64_t prefetch;
2246 struct ib_counters {
2247 struct ib_device *device;
2248 struct ib_uobject *uobject;
2249 /* num of objects attached */
2253 struct ib_counters_read_attr {
2256 u32 flags; /* use enum ib_read_counters_flags */
2259 struct uverbs_attr_bundle;
2261 struct iw_cm_conn_param;
2263 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2264 .size_##ib_struct = \
2265 (sizeof(struct drv_struct) + \
2266 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2267 BUILD_BUG_ON_ZERO( \
2268 !__same_type(((struct drv_struct *)NULL)->member, \
2271 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2272 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2275 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2276 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2279 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2280 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2282 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2284 struct rdma_user_mmap_entry {
2286 struct ib_ucontext *ucontext;
2287 unsigned long start_pgoff;
2289 bool driver_removed;
2292 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2294 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2296 return (u64)entry->start_pgoff << PAGE_SHIFT;
2300 * struct ib_device_ops - InfiniBand device operations
2301 * This structure defines all the InfiniBand device operations, providers will
2302 * need to define the supported operations, otherwise they will be set to null.
2304 struct ib_device_ops {
2305 struct module *owner;
2306 enum rdma_driver_id driver_id;
2308 unsigned int uverbs_no_driver_id_binding:1;
2311 * NOTE: New drivers should not make use of device_group; instead new
2312 * device parameter should be exposed via netlink command. This
2313 * mechanism exists only for existing drivers.
2315 const struct attribute_group *device_group;
2316 const struct attribute_group **port_groups;
2318 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2319 const struct ib_send_wr **bad_send_wr);
2320 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2321 const struct ib_recv_wr **bad_recv_wr);
2322 void (*drain_rq)(struct ib_qp *qp);
2323 void (*drain_sq)(struct ib_qp *qp);
2324 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2325 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2326 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2327 int (*post_srq_recv)(struct ib_srq *srq,
2328 const struct ib_recv_wr *recv_wr,
2329 const struct ib_recv_wr **bad_recv_wr);
2330 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2331 u32 port_num, const struct ib_wc *in_wc,
2332 const struct ib_grh *in_grh,
2333 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2334 size_t *out_mad_size, u16 *out_mad_pkey_index);
2335 int (*query_device)(struct ib_device *device,
2336 struct ib_device_attr *device_attr,
2337 struct ib_udata *udata);
2338 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2339 struct ib_device_modify *device_modify);
2340 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2341 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2343 int (*query_port)(struct ib_device *device, u32 port_num,
2344 struct ib_port_attr *port_attr);
2345 int (*modify_port)(struct ib_device *device, u32 port_num,
2346 int port_modify_mask,
2347 struct ib_port_modify *port_modify);
2349 * The following mandatory functions are used only at device
2350 * registration. Keep functions such as these at the end of this
2351 * structure to avoid cache line misses when accessing struct ib_device
2354 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2355 struct ib_port_immutable *immutable);
2356 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2359 * When calling get_netdev, the HW vendor's driver should return the
2360 * net device of device @device at port @port_num or NULL if such
2361 * a net device doesn't exist. The vendor driver should call dev_hold
2362 * on this net device. The HW vendor's device driver must guarantee
2363 * that this function returns NULL before the net device has finished
2364 * NETDEV_UNREGISTER state.
2366 struct net_device *(*get_netdev)(struct ib_device *device,
2369 * rdma netdev operation
2371 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2372 * must return -EOPNOTSUPP if it doesn't support the specified type.
2374 struct net_device *(*alloc_rdma_netdev)(
2375 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2376 const char *name, unsigned char name_assign_type,
2377 void (*setup)(struct net_device *));
2379 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2380 enum rdma_netdev_t type,
2381 struct rdma_netdev_alloc_params *params);
2383 * query_gid should be return GID value for @device, when @port_num
2384 * link layer is either IB or iWarp. It is no-op if @port_num port
2385 * is RoCE link layer.
2387 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2390 * When calling add_gid, the HW vendor's driver should add the gid
2391 * of device of port at gid index available at @attr. Meta-info of
2392 * that gid (for example, the network device related to this gid) is
2393 * available at @attr. @context allows the HW vendor driver to store
2394 * extra information together with a GID entry. The HW vendor driver may
2395 * allocate memory to contain this information and store it in @context
2396 * when a new GID entry is written to. Params are consistent until the
2397 * next call of add_gid or delete_gid. The function should return 0 on
2398 * success or error otherwise. The function could be called
2399 * concurrently for different ports. This function is only called when
2400 * roce_gid_table is used.
2402 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2404 * When calling del_gid, the HW vendor's driver should delete the
2405 * gid of device @device at gid index gid_index of port port_num
2406 * available in @attr.
2407 * Upon the deletion of a GID entry, the HW vendor must free any
2408 * allocated memory. The caller will clear @context afterwards.
2409 * This function is only called when roce_gid_table is used.
2411 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2412 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2414 int (*alloc_ucontext)(struct ib_ucontext *context,
2415 struct ib_udata *udata);
2416 void (*dealloc_ucontext)(struct ib_ucontext *context);
2417 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2419 * This will be called once refcount of an entry in mmap_xa reaches
2420 * zero. The type of the memory that was mapped may differ between
2421 * entries and is opaque to the rdma_user_mmap interface.
2422 * Therefore needs to be implemented by the driver in mmap_free.
2424 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2425 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2426 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2427 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2428 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2429 struct ib_udata *udata);
2430 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2431 struct ib_udata *udata);
2432 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2433 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2434 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2435 int (*create_srq)(struct ib_srq *srq,
2436 struct ib_srq_init_attr *srq_init_attr,
2437 struct ib_udata *udata);
2438 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2439 enum ib_srq_attr_mask srq_attr_mask,
2440 struct ib_udata *udata);
2441 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2442 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2443 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2444 struct ib_udata *udata);
2445 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2446 int qp_attr_mask, struct ib_udata *udata);
2447 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2448 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2449 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2450 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2451 struct ib_udata *udata);
2452 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2453 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2454 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2455 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2456 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2457 u64 virt_addr, int mr_access_flags,
2458 struct ib_udata *udata);
2459 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2460 u64 length, u64 virt_addr, int fd,
2461 int mr_access_flags,
2462 struct ib_udata *udata);
2463 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2464 u64 length, u64 virt_addr,
2465 int mr_access_flags, struct ib_pd *pd,
2466 struct ib_udata *udata);
2467 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2468 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2470 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2471 u32 max_num_data_sg,
2472 u32 max_num_meta_sg);
2473 int (*advise_mr)(struct ib_pd *pd,
2474 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2475 struct ib_sge *sg_list, u32 num_sge,
2476 struct uverbs_attr_bundle *attrs);
2479 * Kernel users should universally support relaxed ordering (RO), as
2480 * they are designed to read data only after observing the CQE and use
2481 * the DMA API correctly.
2483 * Some drivers implicitly enable RO if platform supports it.
2485 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2486 unsigned int *sg_offset);
2487 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2488 struct ib_mr_status *mr_status);
2489 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2490 int (*dealloc_mw)(struct ib_mw *mw);
2491 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2492 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2493 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2494 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2495 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2496 struct ib_flow_attr *flow_attr,
2497 struct ib_udata *udata);
2498 int (*destroy_flow)(struct ib_flow *flow_id);
2499 struct ib_flow_action *(*create_flow_action_esp)(
2500 struct ib_device *device,
2501 const struct ib_flow_action_attrs_esp *attr,
2502 struct uverbs_attr_bundle *attrs);
2503 int (*destroy_flow_action)(struct ib_flow_action *action);
2504 int (*modify_flow_action_esp)(
2505 struct ib_flow_action *action,
2506 const struct ib_flow_action_attrs_esp *attr,
2507 struct uverbs_attr_bundle *attrs);
2508 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2510 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2511 struct ifla_vf_info *ivf);
2512 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2513 struct ifla_vf_stats *stats);
2514 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2515 struct ifla_vf_guid *node_guid,
2516 struct ifla_vf_guid *port_guid);
2517 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2519 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2520 struct ib_wq_init_attr *init_attr,
2521 struct ib_udata *udata);
2522 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2523 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2524 u32 wq_attr_mask, struct ib_udata *udata);
2525 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2526 struct ib_rwq_ind_table_init_attr *init_attr,
2527 struct ib_udata *udata);
2528 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2529 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2530 struct ib_ucontext *context,
2531 struct ib_dm_alloc_attr *attr,
2532 struct uverbs_attr_bundle *attrs);
2533 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2534 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2535 struct ib_dm_mr_attr *attr,
2536 struct uverbs_attr_bundle *attrs);
2537 int (*create_counters)(struct ib_counters *counters,
2538 struct uverbs_attr_bundle *attrs);
2539 int (*destroy_counters)(struct ib_counters *counters);
2540 int (*read_counters)(struct ib_counters *counters,
2541 struct ib_counters_read_attr *counters_read_attr,
2542 struct uverbs_attr_bundle *attrs);
2543 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2544 int data_sg_nents, unsigned int *data_sg_offset,
2545 struct scatterlist *meta_sg, int meta_sg_nents,
2546 unsigned int *meta_sg_offset);
2549 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2550 * fill in the driver initialized data. The struct is kfree()'ed by
2551 * the sysfs core when the device is removed. A lifespan of -1 in the
2552 * return struct tells the core to set a default lifespan.
2554 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2555 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2558 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2559 * @index - The index in the value array we wish to have updated, or
2560 * num_counters if we want all stats updated
2562 * < 0 - Error, no counters updated
2563 * index - Updated the single counter pointed to by index
2564 * num_counters - Updated all counters (will reset the timestamp
2565 * and prevent further calls for lifespan milliseconds)
2566 * Drivers are allowed to update all counters in leiu of just the
2567 * one given in index at their option
2569 int (*get_hw_stats)(struct ib_device *device,
2570 struct rdma_hw_stats *stats, u32 port, int index);
2573 * Allows rdma drivers to add their own restrack attributes.
2575 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2576 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2577 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2578 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2579 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2580 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2581 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2583 /* Device lifecycle callbacks */
2585 * Called after the device becomes registered, before clients are
2588 int (*enable_driver)(struct ib_device *dev);
2590 * This is called as part of ib_dealloc_device().
2592 void (*dealloc_driver)(struct ib_device *dev);
2594 /* iWarp CM callbacks */
2595 void (*iw_add_ref)(struct ib_qp *qp);
2596 void (*iw_rem_ref)(struct ib_qp *qp);
2597 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2598 int (*iw_connect)(struct iw_cm_id *cm_id,
2599 struct iw_cm_conn_param *conn_param);
2600 int (*iw_accept)(struct iw_cm_id *cm_id,
2601 struct iw_cm_conn_param *conn_param);
2602 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2604 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2605 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2607 * counter_bind_qp - Bind a QP to a counter.
2608 * @counter - The counter to be bound. If counter->id is zero then
2609 * the driver needs to allocate a new counter and set counter->id
2611 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2613 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2614 * counter and bind it onto the default one
2616 int (*counter_unbind_qp)(struct ib_qp *qp);
2618 * counter_dealloc -De-allocate the hw counter
2620 int (*counter_dealloc)(struct rdma_counter *counter);
2622 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2623 * the driver initialized data.
2625 struct rdma_hw_stats *(*counter_alloc_stats)(
2626 struct rdma_counter *counter);
2628 * counter_update_stats - Query the stats value of this counter
2630 int (*counter_update_stats)(struct rdma_counter *counter);
2633 * Allows rdma drivers to add their own restrack attributes
2634 * dumped via 'rdma stat' iproute2 command.
2636 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2638 /* query driver for its ucontext properties */
2639 int (*query_ucontext)(struct ib_ucontext *context,
2640 struct uverbs_attr_bundle *attrs);
2643 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2644 * Everyone else relies on Linux memory management model.
2646 int (*get_numa_node)(struct ib_device *dev);
2648 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2649 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2650 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2651 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2652 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2653 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2654 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2655 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2656 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2657 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2660 struct ib_core_device {
2661 /* device must be the first element in structure until,
2662 * union of ib_core_device and device exists in ib_device.
2665 possible_net_t rdma_net;
2666 struct kobject *ports_kobj;
2667 struct list_head port_list;
2668 struct ib_device *owner; /* reach back to owner ib_device */
2671 struct rdma_restrack_root;
2673 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2674 struct device *dma_device;
2675 struct ib_device_ops ops;
2676 char name[IB_DEVICE_NAME_MAX];
2677 struct rcu_head rcu_head;
2679 struct list_head event_handler_list;
2680 /* Protects event_handler_list */
2681 struct rw_semaphore event_handler_rwsem;
2683 /* Protects QP's event_handler calls and open_qp list */
2684 spinlock_t qp_open_list_lock;
2686 struct rw_semaphore client_data_rwsem;
2687 struct xarray client_data;
2688 struct mutex unregistration_lock;
2690 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2691 rwlock_t cache_lock;
2693 * port_data is indexed by port number
2695 struct ib_port_data *port_data;
2697 int num_comp_vectors;
2701 struct ib_core_device coredev;
2704 /* First group is for device attributes,
2705 * Second group is for driver provided attributes (optional).
2706 * Third group is for the hw_stats
2707 * It is a NULL terminated array.
2709 const struct attribute_group *groups[4];
2711 u64 uverbs_cmd_mask;
2713 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2717 /* Indicates kernel verbs support, should not be used in drivers */
2718 u16 kverbs_provider:1;
2719 /* CQ adaptive moderation (RDMA DIM) */
2723 struct ib_device_attr attrs;
2724 struct hw_stats_device_data *hw_stats_data;
2726 #ifdef CONFIG_CGROUP_RDMA
2727 struct rdmacg_device cg_device;
2732 spinlock_t cq_pools_lock;
2733 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2735 struct rdma_restrack_root *res;
2737 const struct uapi_definition *driver_def;
2740 * Positive refcount indicates that the device is currently
2741 * registered and cannot be unregistered.
2743 refcount_t refcount;
2744 struct completion unreg_completion;
2745 struct work_struct unregistration_work;
2747 const struct rdma_link_ops *link_ops;
2749 /* Protects compat_devs xarray modifications */
2750 struct mutex compat_devs_mutex;
2751 /* Maintains compat devices for each net namespace */
2752 struct xarray compat_devs;
2754 /* Used by iWarp CM */
2755 char iw_ifname[IFNAMSIZ];
2756 u32 iw_driver_flags;
2760 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2761 gfp_t gfp, bool is_numa_aware)
2763 if (is_numa_aware && dev->ops.get_numa_node)
2764 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2766 return kzalloc(size, gfp);
2769 struct ib_client_nl_info;
2772 int (*add)(struct ib_device *ibdev);
2773 void (*remove)(struct ib_device *, void *client_data);
2774 void (*rename)(struct ib_device *dev, void *client_data);
2775 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2776 struct ib_client_nl_info *res);
2777 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2779 /* Returns the net_dev belonging to this ib_client and matching the
2781 * @dev: An RDMA device that the net_dev use for communication.
2782 * @port: A physical port number on the RDMA device.
2783 * @pkey: P_Key that the net_dev uses if applicable.
2784 * @gid: A GID that the net_dev uses to communicate.
2785 * @addr: An IP address the net_dev is configured with.
2786 * @client_data: The device's client data set by ib_set_client_data().
2788 * An ib_client that implements a net_dev on top of RDMA devices
2789 * (such as IP over IB) should implement this callback, allowing the
2790 * rdma_cm module to find the right net_dev for a given request.
2792 * The caller is responsible for calling dev_put on the returned
2794 struct net_device *(*get_net_dev_by_params)(
2795 struct ib_device *dev,
2798 const union ib_gid *gid,
2799 const struct sockaddr *addr,
2803 struct completion uses_zero;
2806 /* kverbs are not required by the client */
2811 * IB block DMA iterator
2813 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2814 * to a HW supported page size.
2816 struct ib_block_iter {
2817 /* internal states */
2818 struct scatterlist *__sg; /* sg holding the current aligned block */
2819 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2820 unsigned int __sg_nents; /* number of SG entries */
2821 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2822 unsigned int __pg_bit; /* alignment of current block */
2825 struct ib_device *_ib_alloc_device(size_t size);
2826 #define ib_alloc_device(drv_struct, member) \
2827 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2828 BUILD_BUG_ON_ZERO(offsetof( \
2829 struct drv_struct, member))), \
2830 struct drv_struct, member)
2832 void ib_dealloc_device(struct ib_device *device);
2834 void ib_get_device_fw_str(struct ib_device *device, char *str);
2836 int ib_register_device(struct ib_device *device, const char *name,
2837 struct device *dma_device);
2838 void ib_unregister_device(struct ib_device *device);
2839 void ib_unregister_driver(enum rdma_driver_id driver_id);
2840 void ib_unregister_device_and_put(struct ib_device *device);
2841 void ib_unregister_device_queued(struct ib_device *ib_dev);
2843 int ib_register_client (struct ib_client *client);
2844 void ib_unregister_client(struct ib_client *client);
2846 void __rdma_block_iter_start(struct ib_block_iter *biter,
2847 struct scatterlist *sglist,
2849 unsigned long pgsz);
2850 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2853 * rdma_block_iter_dma_address - get the aligned dma address of the current
2854 * block held by the block iterator.
2855 * @biter: block iterator holding the memory block
2857 static inline dma_addr_t
2858 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2860 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2864 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2865 * @sglist: sglist to iterate over
2866 * @biter: block iterator holding the memory block
2867 * @nents: maximum number of sg entries to iterate over
2868 * @pgsz: best HW supported page size to use
2870 * Callers may use rdma_block_iter_dma_address() to get each
2871 * blocks aligned DMA address.
2873 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2874 for (__rdma_block_iter_start(biter, sglist, nents, \
2876 __rdma_block_iter_next(biter);)
2879 * ib_get_client_data - Get IB client context
2880 * @device:Device to get context for
2881 * @client:Client to get context for
2883 * ib_get_client_data() returns the client context data set with
2884 * ib_set_client_data(). This can only be called while the client is
2885 * registered to the device, once the ib_client remove() callback returns this
2888 static inline void *ib_get_client_data(struct ib_device *device,
2889 struct ib_client *client)
2891 return xa_load(&device->client_data, client->client_id);
2893 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2895 void ib_set_device_ops(struct ib_device *device,
2896 const struct ib_device_ops *ops);
2898 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2899 unsigned long pfn, unsigned long size, pgprot_t prot,
2900 struct rdma_user_mmap_entry *entry);
2901 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2902 struct rdma_user_mmap_entry *entry,
2904 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2905 struct rdma_user_mmap_entry *entry,
2906 size_t length, u32 min_pgoff,
2909 struct rdma_user_mmap_entry *
2910 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2911 unsigned long pgoff);
2912 struct rdma_user_mmap_entry *
2913 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2914 struct vm_area_struct *vma);
2915 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2917 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2919 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2921 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2924 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2926 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2929 static inline bool ib_is_buffer_cleared(const void __user *p,
2935 if (len > USHRT_MAX)
2938 buf = memdup_user(p, len);
2942 ret = !memchr_inv(buf, 0, len);
2947 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2951 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2955 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2956 * contains all required attributes and no attributes not allowed for
2957 * the given QP state transition.
2958 * @cur_state: Current QP state
2959 * @next_state: Next QP state
2961 * @mask: Mask of supplied QP attributes
2963 * This function is a helper function that a low-level driver's
2964 * modify_qp method can use to validate the consumer's input. It
2965 * checks that cur_state and next_state are valid QP states, that a
2966 * transition from cur_state to next_state is allowed by the IB spec,
2967 * and that the attribute mask supplied is allowed for the transition.
2969 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2970 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2972 void ib_register_event_handler(struct ib_event_handler *event_handler);
2973 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2974 void ib_dispatch_event(const struct ib_event *event);
2976 int ib_query_port(struct ib_device *device,
2977 u32 port_num, struct ib_port_attr *port_attr);
2979 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2983 * rdma_cap_ib_switch - Check if the device is IB switch
2984 * @device: Device to check
2986 * Device driver is responsible for setting is_switch bit on
2987 * in ib_device structure at init time.
2989 * Return: true if the device is IB switch.
2991 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2993 return device->is_switch;
2997 * rdma_start_port - Return the first valid port number for the device
3000 * @device: Device to be checked
3002 * Return start port number
3004 static inline u32 rdma_start_port(const struct ib_device *device)
3006 return rdma_cap_ib_switch(device) ? 0 : 1;
3010 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3011 * @device - The struct ib_device * to iterate over
3012 * @iter - The unsigned int to store the port number
3014 #define rdma_for_each_port(device, iter) \
3015 for (iter = rdma_start_port(device + \
3016 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3018 iter <= rdma_end_port(device); iter++)
3021 * rdma_end_port - Return the last valid port number for the device
3024 * @device: Device to be checked
3026 * Return last port number
3028 static inline u32 rdma_end_port(const struct ib_device *device)
3030 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3033 static inline int rdma_is_port_valid(const struct ib_device *device,
3036 return (port >= rdma_start_port(device) &&
3037 port <= rdma_end_port(device));
3040 static inline bool rdma_is_grh_required(const struct ib_device *device,
3043 return device->port_data[port_num].immutable.core_cap_flags &
3044 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3047 static inline bool rdma_protocol_ib(const struct ib_device *device,
3050 return device->port_data[port_num].immutable.core_cap_flags &
3051 RDMA_CORE_CAP_PROT_IB;
3054 static inline bool rdma_protocol_roce(const struct ib_device *device,
3057 return device->port_data[port_num].immutable.core_cap_flags &
3058 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3061 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3064 return device->port_data[port_num].immutable.core_cap_flags &
3065 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3068 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3071 return device->port_data[port_num].immutable.core_cap_flags &
3072 RDMA_CORE_CAP_PROT_ROCE;
3075 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3078 return device->port_data[port_num].immutable.core_cap_flags &
3079 RDMA_CORE_CAP_PROT_IWARP;
3082 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3085 return rdma_protocol_ib(device, port_num) ||
3086 rdma_protocol_roce(device, port_num);
3089 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3092 return device->port_data[port_num].immutable.core_cap_flags &
3093 RDMA_CORE_CAP_PROT_RAW_PACKET;
3096 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3099 return device->port_data[port_num].immutable.core_cap_flags &
3100 RDMA_CORE_CAP_PROT_USNIC;
3104 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3105 * Management Datagrams.
3106 * @device: Device to check
3107 * @port_num: Port number to check
3109 * Management Datagrams (MAD) are a required part of the InfiniBand
3110 * specification and are supported on all InfiniBand devices. A slightly
3111 * extended version are also supported on OPA interfaces.
3113 * Return: true if the port supports sending/receiving of MAD packets.
3115 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3117 return device->port_data[port_num].immutable.core_cap_flags &
3118 RDMA_CORE_CAP_IB_MAD;
3122 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3123 * Management Datagrams.
3124 * @device: Device to check
3125 * @port_num: Port number to check
3127 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3128 * datagrams with their own versions. These OPA MADs share many but not all of
3129 * the characteristics of InfiniBand MADs.
3131 * OPA MADs differ in the following ways:
3133 * 1) MADs are variable size up to 2K
3134 * IBTA defined MADs remain fixed at 256 bytes
3135 * 2) OPA SMPs must carry valid PKeys
3136 * 3) OPA SMP packets are a different format
3138 * Return: true if the port supports OPA MAD packet formats.
3140 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3142 return device->port_data[port_num].immutable.core_cap_flags &
3143 RDMA_CORE_CAP_OPA_MAD;
3147 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3148 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3149 * @device: Device to check
3150 * @port_num: Port number to check
3152 * Each InfiniBand node is required to provide a Subnet Management Agent
3153 * that the subnet manager can access. Prior to the fabric being fully
3154 * configured by the subnet manager, the SMA is accessed via a well known
3155 * interface called the Subnet Management Interface (SMI). This interface
3156 * uses directed route packets to communicate with the SM to get around the
3157 * chicken and egg problem of the SM needing to know what's on the fabric
3158 * in order to configure the fabric, and needing to configure the fabric in
3159 * order to send packets to the devices on the fabric. These directed
3160 * route packets do not need the fabric fully configured in order to reach
3161 * their destination. The SMI is the only method allowed to send
3162 * directed route packets on an InfiniBand fabric.
3164 * Return: true if the port provides an SMI.
3166 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3168 return device->port_data[port_num].immutable.core_cap_flags &
3169 RDMA_CORE_CAP_IB_SMI;
3173 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3174 * Communication Manager.
3175 * @device: Device to check
3176 * @port_num: Port number to check
3178 * The InfiniBand Communication Manager is one of many pre-defined General
3179 * Service Agents (GSA) that are accessed via the General Service
3180 * Interface (GSI). It's role is to facilitate establishment of connections
3181 * between nodes as well as other management related tasks for established
3184 * Return: true if the port supports an IB CM (this does not guarantee that
3185 * a CM is actually running however).
3187 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3189 return device->port_data[port_num].immutable.core_cap_flags &
3190 RDMA_CORE_CAP_IB_CM;
3194 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3195 * Communication Manager.
3196 * @device: Device to check
3197 * @port_num: Port number to check
3199 * Similar to above, but specific to iWARP connections which have a different
3200 * managment protocol than InfiniBand.
3202 * Return: true if the port supports an iWARP CM (this does not guarantee that
3203 * a CM is actually running however).
3205 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3207 return device->port_data[port_num].immutable.core_cap_flags &
3208 RDMA_CORE_CAP_IW_CM;
3212 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3213 * Subnet Administration.
3214 * @device: Device to check
3215 * @port_num: Port number to check
3217 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3218 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3219 * fabrics, devices should resolve routes to other hosts by contacting the
3220 * SA to query the proper route.
3222 * Return: true if the port should act as a client to the fabric Subnet
3223 * Administration interface. This does not imply that the SA service is
3226 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3228 return device->port_data[port_num].immutable.core_cap_flags &
3229 RDMA_CORE_CAP_IB_SA;
3233 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3235 * @device: Device to check
3236 * @port_num: Port number to check
3238 * InfiniBand multicast registration is more complex than normal IPv4 or
3239 * IPv6 multicast registration. Each Host Channel Adapter must register
3240 * with the Subnet Manager when it wishes to join a multicast group. It
3241 * should do so only once regardless of how many queue pairs it subscribes
3242 * to this group. And it should leave the group only after all queue pairs
3243 * attached to the group have been detached.
3245 * Return: true if the port must undertake the additional adminstrative
3246 * overhead of registering/unregistering with the SM and tracking of the
3247 * total number of queue pairs attached to the multicast group.
3249 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3252 return rdma_cap_ib_sa(device, port_num);
3256 * rdma_cap_af_ib - Check if the port of device has the capability
3257 * Native Infiniband Address.
3258 * @device: Device to check
3259 * @port_num: Port number to check
3261 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3262 * GID. RoCE uses a different mechanism, but still generates a GID via
3263 * a prescribed mechanism and port specific data.
3265 * Return: true if the port uses a GID address to identify devices on the
3268 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3270 return device->port_data[port_num].immutable.core_cap_flags &
3271 RDMA_CORE_CAP_AF_IB;
3275 * rdma_cap_eth_ah - Check if the port of device has the capability
3276 * Ethernet Address Handle.
3277 * @device: Device to check
3278 * @port_num: Port number to check
3280 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3281 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3282 * port. Normally, packet headers are generated by the sending host
3283 * adapter, but when sending connectionless datagrams, we must manually
3284 * inject the proper headers for the fabric we are communicating over.
3286 * Return: true if we are running as a RoCE port and must force the
3287 * addition of a Global Route Header built from our Ethernet Address
3288 * Handle into our header list for connectionless packets.
3290 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3292 return device->port_data[port_num].immutable.core_cap_flags &
3293 RDMA_CORE_CAP_ETH_AH;
3297 * rdma_cap_opa_ah - Check if the port of device supports
3298 * OPA Address handles
3299 * @device: Device to check
3300 * @port_num: Port number to check
3302 * Return: true if we are running on an OPA device which supports
3303 * the extended OPA addressing.
3305 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3307 return (device->port_data[port_num].immutable.core_cap_flags &
3308 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3312 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3315 * @port_num: Port number
3317 * This MAD size includes the MAD headers and MAD payload. No other headers
3320 * Return the max MAD size required by the Port. Will return 0 if the port
3321 * does not support MADs
3323 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3326 return device->port_data[port_num].immutable.max_mad_size;
3330 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3331 * @device: Device to check
3332 * @port_num: Port number to check
3334 * RoCE GID table mechanism manages the various GIDs for a device.
3336 * NOTE: if allocating the port's GID table has failed, this call will still
3337 * return true, but any RoCE GID table API will fail.
3339 * Return: true if the port uses RoCE GID table mechanism in order to manage
3342 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3345 return rdma_protocol_roce(device, port_num) &&
3346 device->ops.add_gid && device->ops.del_gid;
3350 * Check if the device supports READ W/ INVALIDATE.
3352 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3355 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3356 * has support for it yet.
3358 return rdma_protocol_iwarp(dev, port_num);
3362 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3364 * @port_num: 1 based Port number
3366 * Return true if port is an Intel OPA port , false if not
3368 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3371 return (device->port_data[port_num].immutable.core_cap_flags &
3372 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3376 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3378 * @port_num: Port number
3379 * @mtu: enum value of MTU
3381 * Return the MTU size supported by the port as an integer value. Will return
3382 * -1 if enum value of mtu is not supported.
3384 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3387 if (rdma_core_cap_opa_port(device, port))
3388 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3390 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3394 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3396 * @port_num: Port number
3397 * @attr: port attribute
3399 * Return the MTU size supported by the port as an integer value.
3401 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3402 struct ib_port_attr *attr)
3404 if (rdma_core_cap_opa_port(device, port))
3405 return attr->phys_mtu;
3407 return ib_mtu_enum_to_int(attr->max_mtu);
3410 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3412 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3413 struct ifla_vf_info *info);
3414 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3415 struct ifla_vf_stats *stats);
3416 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3417 struct ifla_vf_guid *node_guid,
3418 struct ifla_vf_guid *port_guid);
3419 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3422 int ib_query_pkey(struct ib_device *device,
3423 u32 port_num, u16 index, u16 *pkey);
3425 int ib_modify_device(struct ib_device *device,
3426 int device_modify_mask,
3427 struct ib_device_modify *device_modify);
3429 int ib_modify_port(struct ib_device *device,
3430 u32 port_num, int port_modify_mask,
3431 struct ib_port_modify *port_modify);
3433 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3434 u32 *port_num, u16 *index);
3436 int ib_find_pkey(struct ib_device *device,
3437 u32 port_num, u16 pkey, u16 *index);
3441 * Create a memory registration for all memory in the system and place
3442 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3443 * ULPs to avoid the overhead of dynamic MRs.
3445 * This flag is generally considered unsafe and must only be used in
3446 * extremly trusted environments. Every use of it will log a warning
3447 * in the kernel log.
3449 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3452 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3453 const char *caller);
3456 * ib_alloc_pd - Allocates an unused protection domain.
3457 * @device: The device on which to allocate the protection domain.
3458 * @flags: protection domain flags
3460 * A protection domain object provides an association between QPs, shared
3461 * receive queues, address handles, memory regions, and memory windows.
3463 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3464 * memory operations.
3466 #define ib_alloc_pd(device, flags) \
3467 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3469 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3472 * ib_dealloc_pd - Deallocate kernel PD
3473 * @pd: The protection domain
3475 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3477 static inline void ib_dealloc_pd(struct ib_pd *pd)
3479 int ret = ib_dealloc_pd_user(pd, NULL);
3481 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3484 enum rdma_create_ah_flags {
3485 /* In a sleepable context */
3486 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3490 * rdma_create_ah - Creates an address handle for the given address vector.
3491 * @pd: The protection domain associated with the address handle.
3492 * @ah_attr: The attributes of the address vector.
3493 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3495 * The address handle is used to reference a local or global destination
3496 * in all UD QP post sends.
3498 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3502 * rdma_create_user_ah - Creates an address handle for the given address vector.
3503 * It resolves destination mac address for ah attribute of RoCE type.
3504 * @pd: The protection domain associated with the address handle.
3505 * @ah_attr: The attributes of the address vector.
3506 * @udata: pointer to user's input output buffer information need by
3509 * It returns 0 on success and returns appropriate error code on error.
3510 * The address handle is used to reference a local or global destination
3511 * in all UD QP post sends.
3513 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3514 struct rdma_ah_attr *ah_attr,
3515 struct ib_udata *udata);
3517 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3519 * @hdr: the L3 header to parse
3520 * @net_type: type of header to parse
3521 * @sgid: place to store source gid
3522 * @dgid: place to store destination gid
3524 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3525 enum rdma_network_type net_type,
3526 union ib_gid *sgid, union ib_gid *dgid);
3529 * ib_get_rdma_header_version - Get the header version
3530 * @hdr: the L3 header to parse
3532 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3535 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3537 * @device: Device on which the received message arrived.
3538 * @port_num: Port on which the received message arrived.
3539 * @wc: Work completion associated with the received message.
3540 * @grh: References the received global route header. This parameter is
3541 * ignored unless the work completion indicates that the GRH is valid.
3542 * @ah_attr: Returned attributes that can be used when creating an address
3543 * handle for replying to the message.
3544 * When ib_init_ah_attr_from_wc() returns success,
3545 * (a) for IB link layer it optionally contains a reference to SGID attribute
3546 * when GRH is present for IB link layer.
3547 * (b) for RoCE link layer it contains a reference to SGID attribute.
3548 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3549 * attributes which are initialized using ib_init_ah_attr_from_wc().
3552 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3553 const struct ib_wc *wc, const struct ib_grh *grh,
3554 struct rdma_ah_attr *ah_attr);
3557 * ib_create_ah_from_wc - Creates an address handle associated with the
3558 * sender of the specified work completion.
3559 * @pd: The protection domain associated with the address handle.
3560 * @wc: Work completion information associated with a received message.
3561 * @grh: References the received global route header. This parameter is
3562 * ignored unless the work completion indicates that the GRH is valid.
3563 * @port_num: The outbound port number to associate with the address.
3565 * The address handle is used to reference a local or global destination
3566 * in all UD QP post sends.
3568 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3569 const struct ib_grh *grh, u32 port_num);
3572 * rdma_modify_ah - Modifies the address vector associated with an address
3574 * @ah: The address handle to modify.
3575 * @ah_attr: The new address vector attributes to associate with the
3578 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3581 * rdma_query_ah - Queries the address vector associated with an address
3583 * @ah: The address handle to query.
3584 * @ah_attr: The address vector attributes associated with the address
3587 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3589 enum rdma_destroy_ah_flags {
3590 /* In a sleepable context */
3591 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3595 * rdma_destroy_ah_user - Destroys an address handle.
3596 * @ah: The address handle to destroy.
3597 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3598 * @udata: Valid user data or NULL for kernel objects
3600 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3603 * rdma_destroy_ah - Destroys an kernel address handle.
3604 * @ah: The address handle to destroy.
3605 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3607 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3609 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3611 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3613 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3616 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3617 struct ib_srq_init_attr *srq_init_attr,
3618 struct ib_usrq_object *uobject,
3619 struct ib_udata *udata);
3620 static inline struct ib_srq *
3621 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3623 if (!pd->device->ops.create_srq)
3624 return ERR_PTR(-EOPNOTSUPP);
3626 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3630 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3631 * @srq: The SRQ to modify.
3632 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3633 * the current values of selected SRQ attributes are returned.
3634 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3635 * are being modified.
3637 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3638 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3639 * the number of receives queued drops below the limit.
3641 int ib_modify_srq(struct ib_srq *srq,
3642 struct ib_srq_attr *srq_attr,
3643 enum ib_srq_attr_mask srq_attr_mask);
3646 * ib_query_srq - Returns the attribute list and current values for the
3648 * @srq: The SRQ to query.
3649 * @srq_attr: The attributes of the specified SRQ.
3651 int ib_query_srq(struct ib_srq *srq,
3652 struct ib_srq_attr *srq_attr);
3655 * ib_destroy_srq_user - Destroys the specified SRQ.
3656 * @srq: The SRQ to destroy.
3657 * @udata: Valid user data or NULL for kernel objects
3659 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3662 * ib_destroy_srq - Destroys the specified kernel SRQ.
3663 * @srq: The SRQ to destroy.
3665 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3667 static inline void ib_destroy_srq(struct ib_srq *srq)
3669 int ret = ib_destroy_srq_user(srq, NULL);
3671 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3675 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3676 * @srq: The SRQ to post the work request on.
3677 * @recv_wr: A list of work requests to post on the receive queue.
3678 * @bad_recv_wr: On an immediate failure, this parameter will reference
3679 * the work request that failed to be posted on the QP.
3681 static inline int ib_post_srq_recv(struct ib_srq *srq,
3682 const struct ib_recv_wr *recv_wr,
3683 const struct ib_recv_wr **bad_recv_wr)
3685 const struct ib_recv_wr *dummy;
3687 return srq->device->ops.post_srq_recv(srq, recv_wr,
3688 bad_recv_wr ? : &dummy);
3691 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3692 struct ib_qp_init_attr *qp_init_attr,
3693 const char *caller);
3695 * ib_create_qp - Creates a kernel QP associated with the specific protection
3697 * @pd: The protection domain associated with the QP.
3698 * @init_attr: A list of initial attributes required to create the
3699 * QP. If QP creation succeeds, then the attributes are updated to
3700 * the actual capabilities of the created QP.
3702 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3703 struct ib_qp_init_attr *init_attr)
3705 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3709 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3710 * @qp: The QP to modify.
3711 * @attr: On input, specifies the QP attributes to modify. On output,
3712 * the current values of selected QP attributes are returned.
3713 * @attr_mask: A bit-mask used to specify which attributes of the QP
3714 * are being modified.
3715 * @udata: pointer to user's input output buffer information
3716 * are being modified.
3717 * It returns 0 on success and returns appropriate error code on error.
3719 int ib_modify_qp_with_udata(struct ib_qp *qp,
3720 struct ib_qp_attr *attr,
3722 struct ib_udata *udata);
3725 * ib_modify_qp - Modifies the attributes for the specified QP and then
3726 * transitions the QP to the given state.
3727 * @qp: The QP to modify.
3728 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3729 * the current values of selected QP attributes are returned.
3730 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3731 * are being modified.
3733 int ib_modify_qp(struct ib_qp *qp,
3734 struct ib_qp_attr *qp_attr,
3738 * ib_query_qp - Returns the attribute list and current values for the
3740 * @qp: The QP to query.
3741 * @qp_attr: The attributes of the specified QP.
3742 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3743 * @qp_init_attr: Additional attributes of the selected QP.
3745 * The qp_attr_mask may be used to limit the query to gathering only the
3746 * selected attributes.
3748 int ib_query_qp(struct ib_qp *qp,
3749 struct ib_qp_attr *qp_attr,
3751 struct ib_qp_init_attr *qp_init_attr);
3754 * ib_destroy_qp - Destroys the specified QP.
3755 * @qp: The QP to destroy.
3756 * @udata: Valid udata or NULL for kernel objects
3758 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3761 * ib_destroy_qp - Destroys the specified kernel QP.
3762 * @qp: The QP to destroy.
3764 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3766 static inline int ib_destroy_qp(struct ib_qp *qp)
3768 return ib_destroy_qp_user(qp, NULL);
3772 * ib_open_qp - Obtain a reference to an existing sharable QP.
3773 * @xrcd - XRC domain
3774 * @qp_open_attr: Attributes identifying the QP to open.
3776 * Returns a reference to a sharable QP.
3778 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3779 struct ib_qp_open_attr *qp_open_attr);
3782 * ib_close_qp - Release an external reference to a QP.
3783 * @qp: The QP handle to release
3785 * The opened QP handle is released by the caller. The underlying
3786 * shared QP is not destroyed until all internal references are released.
3788 int ib_close_qp(struct ib_qp *qp);
3791 * ib_post_send - Posts a list of work requests to the send queue of
3793 * @qp: The QP to post the work request on.
3794 * @send_wr: A list of work requests to post on the send queue.
3795 * @bad_send_wr: On an immediate failure, this parameter will reference
3796 * the work request that failed to be posted on the QP.
3798 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3799 * error is returned, the QP state shall not be affected,
3800 * ib_post_send() will return an immediate error after queueing any
3801 * earlier work requests in the list.
3803 static inline int ib_post_send(struct ib_qp *qp,
3804 const struct ib_send_wr *send_wr,
3805 const struct ib_send_wr **bad_send_wr)
3807 const struct ib_send_wr *dummy;
3809 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3813 * ib_post_recv - Posts a list of work requests to the receive queue of
3815 * @qp: The QP to post the work request on.
3816 * @recv_wr: A list of work requests to post on the receive queue.
3817 * @bad_recv_wr: On an immediate failure, this parameter will reference
3818 * the work request that failed to be posted on the QP.
3820 static inline int ib_post_recv(struct ib_qp *qp,
3821 const struct ib_recv_wr *recv_wr,
3822 const struct ib_recv_wr **bad_recv_wr)
3824 const struct ib_recv_wr *dummy;
3826 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3829 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3830 int comp_vector, enum ib_poll_context poll_ctx,
3831 const char *caller);
3832 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3833 int nr_cqe, int comp_vector,
3834 enum ib_poll_context poll_ctx)
3836 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3840 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3841 int nr_cqe, enum ib_poll_context poll_ctx,
3842 const char *caller);
3845 * ib_alloc_cq_any: Allocate kernel CQ
3846 * @dev: The IB device
3847 * @private: Private data attached to the CQE
3848 * @nr_cqe: Number of CQEs in the CQ
3849 * @poll_ctx: Context used for polling the CQ
3851 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3852 void *private, int nr_cqe,
3853 enum ib_poll_context poll_ctx)
3855 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3859 void ib_free_cq(struct ib_cq *cq);
3860 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3863 * ib_create_cq - Creates a CQ on the specified device.
3864 * @device: The device on which to create the CQ.
3865 * @comp_handler: A user-specified callback that is invoked when a
3866 * completion event occurs on the CQ.
3867 * @event_handler: A user-specified callback that is invoked when an
3868 * asynchronous event not associated with a completion occurs on the CQ.
3869 * @cq_context: Context associated with the CQ returned to the user via
3870 * the associated completion and event handlers.
3871 * @cq_attr: The attributes the CQ should be created upon.
3873 * Users can examine the cq structure to determine the actual CQ size.
3875 struct ib_cq *__ib_create_cq(struct ib_device *device,
3876 ib_comp_handler comp_handler,
3877 void (*event_handler)(struct ib_event *, void *),
3879 const struct ib_cq_init_attr *cq_attr,
3880 const char *caller);
3881 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3882 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3885 * ib_resize_cq - Modifies the capacity of the CQ.
3886 * @cq: The CQ to resize.
3887 * @cqe: The minimum size of the CQ.
3889 * Users can examine the cq structure to determine the actual CQ size.
3891 int ib_resize_cq(struct ib_cq *cq, int cqe);
3894 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3895 * @cq: The CQ to modify.
3896 * @cq_count: number of CQEs that will trigger an event
3897 * @cq_period: max period of time in usec before triggering an event
3900 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3903 * ib_destroy_cq_user - Destroys the specified CQ.
3904 * @cq: The CQ to destroy.
3905 * @udata: Valid user data or NULL for kernel objects
3907 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3910 * ib_destroy_cq - Destroys the specified kernel CQ.
3911 * @cq: The CQ to destroy.
3913 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3915 static inline void ib_destroy_cq(struct ib_cq *cq)
3917 int ret = ib_destroy_cq_user(cq, NULL);
3919 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3923 * ib_poll_cq - poll a CQ for completion(s)
3924 * @cq:the CQ being polled
3925 * @num_entries:maximum number of completions to return
3926 * @wc:array of at least @num_entries &struct ib_wc where completions
3929 * Poll a CQ for (possibly multiple) completions. If the return value
3930 * is < 0, an error occurred. If the return value is >= 0, it is the
3931 * number of completions returned. If the return value is
3932 * non-negative and < num_entries, then the CQ was emptied.
3934 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3937 return cq->device->ops.poll_cq(cq, num_entries, wc);
3941 * ib_req_notify_cq - Request completion notification on a CQ.
3942 * @cq: The CQ to generate an event for.
3944 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3945 * to request an event on the next solicited event or next work
3946 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3947 * may also be |ed in to request a hint about missed events, as
3951 * < 0 means an error occurred while requesting notification
3952 * == 0 means notification was requested successfully, and if
3953 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3954 * were missed and it is safe to wait for another event. In
3955 * this case is it guaranteed that any work completions added
3956 * to the CQ since the last CQ poll will trigger a completion
3957 * notification event.
3958 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3959 * in. It means that the consumer must poll the CQ again to
3960 * make sure it is empty to avoid missing an event because of a
3961 * race between requesting notification and an entry being
3962 * added to the CQ. This return value means it is possible
3963 * (but not guaranteed) that a work completion has been added
3964 * to the CQ since the last poll without triggering a
3965 * completion notification event.
3967 static inline int ib_req_notify_cq(struct ib_cq *cq,
3968 enum ib_cq_notify_flags flags)
3970 return cq->device->ops.req_notify_cq(cq, flags);
3973 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3974 int comp_vector_hint,
3975 enum ib_poll_context poll_ctx);
3977 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3980 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3981 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3982 * address into the dma address.
3984 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3986 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3990 * ib_dma_mapping_error - check a DMA addr for error
3991 * @dev: The device for which the dma_addr was created
3992 * @dma_addr: The DMA address to check
3994 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3996 if (ib_uses_virt_dma(dev))
3998 return dma_mapping_error(dev->dma_device, dma_addr);
4002 * ib_dma_map_single - Map a kernel virtual address to DMA address
4003 * @dev: The device for which the dma_addr is to be created
4004 * @cpu_addr: The kernel virtual address
4005 * @size: The size of the region in bytes
4006 * @direction: The direction of the DMA
4008 static inline u64 ib_dma_map_single(struct ib_device *dev,
4009 void *cpu_addr, size_t size,
4010 enum dma_data_direction direction)
4012 if (ib_uses_virt_dma(dev))
4013 return (uintptr_t)cpu_addr;
4014 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4018 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4019 * @dev: The device for which the DMA address was created
4020 * @addr: The DMA address
4021 * @size: The size of the region in bytes
4022 * @direction: The direction of the DMA
4024 static inline void ib_dma_unmap_single(struct ib_device *dev,
4025 u64 addr, size_t size,
4026 enum dma_data_direction direction)
4028 if (!ib_uses_virt_dma(dev))
4029 dma_unmap_single(dev->dma_device, addr, size, direction);
4033 * ib_dma_map_page - Map a physical page to DMA address
4034 * @dev: The device for which the dma_addr is to be created
4035 * @page: The page to be mapped
4036 * @offset: The offset within the page
4037 * @size: The size of the region in bytes
4038 * @direction: The direction of the DMA
4040 static inline u64 ib_dma_map_page(struct ib_device *dev,
4042 unsigned long offset,
4044 enum dma_data_direction direction)
4046 if (ib_uses_virt_dma(dev))
4047 return (uintptr_t)(page_address(page) + offset);
4048 return dma_map_page(dev->dma_device, page, offset, size, direction);
4052 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4053 * @dev: The device for which the DMA address was created
4054 * @addr: The DMA address
4055 * @size: The size of the region in bytes
4056 * @direction: The direction of the DMA
4058 static inline void ib_dma_unmap_page(struct ib_device *dev,
4059 u64 addr, size_t size,
4060 enum dma_data_direction direction)
4062 if (!ib_uses_virt_dma(dev))
4063 dma_unmap_page(dev->dma_device, addr, size, direction);
4066 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4067 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4068 struct scatterlist *sg, int nents,
4069 enum dma_data_direction direction,
4070 unsigned long dma_attrs)
4072 if (ib_uses_virt_dma(dev))
4073 return ib_dma_virt_map_sg(dev, sg, nents);
4074 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4078 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4079 struct scatterlist *sg, int nents,
4080 enum dma_data_direction direction,
4081 unsigned long dma_attrs)
4083 if (!ib_uses_virt_dma(dev))
4084 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4089 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4090 * @dev: The device for which the DMA addresses are to be created
4091 * @sg: The sg_table object describing the buffer
4092 * @direction: The direction of the DMA
4093 * @attrs: Optional DMA attributes for the map operation
4095 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4096 struct sg_table *sgt,
4097 enum dma_data_direction direction,
4098 unsigned long dma_attrs)
4100 if (ib_uses_virt_dma(dev)) {
4101 ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4104 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4107 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4108 struct sg_table *sgt,
4109 enum dma_data_direction direction,
4110 unsigned long dma_attrs)
4112 if (!ib_uses_virt_dma(dev))
4113 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4117 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4118 * @dev: The device for which the DMA addresses are to be created
4119 * @sg: The array of scatter/gather entries
4120 * @nents: The number of scatter/gather entries
4121 * @direction: The direction of the DMA
4123 static inline int ib_dma_map_sg(struct ib_device *dev,
4124 struct scatterlist *sg, int nents,
4125 enum dma_data_direction direction)
4127 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4131 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4132 * @dev: The device for which the DMA addresses were created
4133 * @sg: The array of scatter/gather entries
4134 * @nents: The number of scatter/gather entries
4135 * @direction: The direction of the DMA
4137 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4138 struct scatterlist *sg, int nents,
4139 enum dma_data_direction direction)
4141 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4145 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4146 * @dev: The device to query
4148 * The returned value represents a size in bytes.
4150 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4152 if (ib_uses_virt_dma(dev))
4154 return dma_get_max_seg_size(dev->dma_device);
4158 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4159 * @dev: The device for which the DMA address was created
4160 * @addr: The DMA address
4161 * @size: The size of the region in bytes
4162 * @dir: The direction of the DMA
4164 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4167 enum dma_data_direction dir)
4169 if (!ib_uses_virt_dma(dev))
4170 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4174 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4175 * @dev: The device for which the DMA address was created
4176 * @addr: The DMA address
4177 * @size: The size of the region in bytes
4178 * @dir: The direction of the DMA
4180 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4183 enum dma_data_direction dir)
4185 if (!ib_uses_virt_dma(dev))
4186 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4189 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4190 * space. This function should be called when 'current' is the owning MM.
4192 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4193 u64 virt_addr, int mr_access_flags);
4195 /* ib_advise_mr - give an advice about an address range in a memory region */
4196 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4197 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4199 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4200 * HCA translation table.
4201 * @mr: The memory region to deregister.
4202 * @udata: Valid user data or NULL for kernel object
4204 * This function can fail, if the memory region has memory windows bound to it.
4206 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4209 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4210 * HCA translation table.
4211 * @mr: The memory region to deregister.
4213 * This function can fail, if the memory region has memory windows bound to it.
4215 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4217 static inline int ib_dereg_mr(struct ib_mr *mr)
4219 return ib_dereg_mr_user(mr, NULL);
4222 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4225 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4226 u32 max_num_data_sg,
4227 u32 max_num_meta_sg);
4230 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4232 * @mr - struct ib_mr pointer to be updated.
4233 * @newkey - new key to be used.
4235 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4237 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4238 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4242 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4243 * for calculating a new rkey for type 2 memory windows.
4244 * @rkey - the rkey to increment.
4246 static inline u32 ib_inc_rkey(u32 rkey)
4248 const u32 mask = 0x000000ff;
4249 return ((rkey + 1) & mask) | (rkey & ~mask);
4253 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4254 * @qp: QP to attach to the multicast group. The QP must be type
4256 * @gid: Multicast group GID.
4257 * @lid: Multicast group LID in host byte order.
4259 * In order to send and receive multicast packets, subnet
4260 * administration must have created the multicast group and configured
4261 * the fabric appropriately. The port associated with the specified
4262 * QP must also be a member of the multicast group.
4264 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4267 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4268 * @qp: QP to detach from the multicast group.
4269 * @gid: Multicast group GID.
4270 * @lid: Multicast group LID in host byte order.
4272 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4274 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4275 struct inode *inode, struct ib_udata *udata);
4276 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4278 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4282 * Local write permission is required if remote write or
4283 * remote atomic permission is also requested.
4285 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4286 !(flags & IB_ACCESS_LOCAL_WRITE))
4289 if (flags & ~IB_ACCESS_SUPPORTED)
4292 if (flags & IB_ACCESS_ON_DEMAND &&
4293 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4298 static inline bool ib_access_writable(int access_flags)
4301 * We have writable memory backing the MR if any of the following
4302 * access flags are set. "Local write" and "remote write" obviously
4303 * require write access. "Remote atomic" can do things like fetch and
4304 * add, which will modify memory, and "MW bind" can change permissions
4305 * by binding a window.
4307 return access_flags &
4308 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4309 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4313 * ib_check_mr_status: lightweight check of MR status.
4314 * This routine may provide status checks on a selected
4315 * ib_mr. first use is for signature status check.
4317 * @mr: A memory region.
4318 * @check_mask: Bitmask of which checks to perform from
4319 * ib_mr_status_check enumeration.
4320 * @mr_status: The container of relevant status checks.
4321 * failed checks will be indicated in the status bitmask
4322 * and the relevant info shall be in the error item.
4324 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4325 struct ib_mr_status *mr_status);
4328 * ib_device_try_get: Hold a registration lock
4329 * device: The device to lock
4331 * A device under an active registration lock cannot become unregistered. It
4332 * is only possible to obtain a registration lock on a device that is fully
4333 * registered, otherwise this function returns false.
4335 * The registration lock is only necessary for actions which require the
4336 * device to still be registered. Uses that only require the device pointer to
4337 * be valid should use get_device(&ibdev->dev) to hold the memory.
4340 static inline bool ib_device_try_get(struct ib_device *dev)
4342 return refcount_inc_not_zero(&dev->refcount);
4345 void ib_device_put(struct ib_device *device);
4346 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4347 enum rdma_driver_id driver_id);
4348 struct ib_device *ib_device_get_by_name(const char *name,
4349 enum rdma_driver_id driver_id);
4350 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4351 u16 pkey, const union ib_gid *gid,
4352 const struct sockaddr *addr);
4353 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4355 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4357 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4358 struct ib_wq_init_attr *init_attr);
4359 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4361 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4362 unsigned int *sg_offset, unsigned int page_size);
4363 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4364 int data_sg_nents, unsigned int *data_sg_offset,
4365 struct scatterlist *meta_sg, int meta_sg_nents,
4366 unsigned int *meta_sg_offset, unsigned int page_size);
4369 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4370 unsigned int *sg_offset, unsigned int page_size)
4374 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4380 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4381 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4383 void ib_drain_rq(struct ib_qp *qp);
4384 void ib_drain_sq(struct ib_qp *qp);
4385 void ib_drain_qp(struct ib_qp *qp);
4387 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4390 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4392 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4393 return attr->roce.dmac;
4397 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4399 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4400 attr->ib.dlid = (u16)dlid;
4401 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4402 attr->opa.dlid = dlid;
4405 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4407 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4408 return attr->ib.dlid;
4409 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4410 return attr->opa.dlid;
4414 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4419 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4424 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4427 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4428 attr->ib.src_path_bits = src_path_bits;
4429 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4430 attr->opa.src_path_bits = src_path_bits;
4433 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4435 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4436 return attr->ib.src_path_bits;
4437 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4438 return attr->opa.src_path_bits;
4442 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4445 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4446 attr->opa.make_grd = make_grd;
4449 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4451 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4452 return attr->opa.make_grd;
4456 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4458 attr->port_num = port_num;
4461 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4463 return attr->port_num;
4466 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4469 attr->static_rate = static_rate;
4472 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4474 return attr->static_rate;
4477 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4478 enum ib_ah_flags flag)
4480 attr->ah_flags = flag;
4483 static inline enum ib_ah_flags
4484 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4486 return attr->ah_flags;
4489 static inline const struct ib_global_route
4490 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4495 /*To retrieve and modify the grh */
4496 static inline struct ib_global_route
4497 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4502 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4504 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4506 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4509 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4512 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4514 grh->dgid.global.subnet_prefix = prefix;
4517 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4520 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4522 grh->dgid.global.interface_id = if_id;
4525 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4526 union ib_gid *dgid, u32 flow_label,
4527 u8 sgid_index, u8 hop_limit,
4530 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4532 attr->ah_flags = IB_AH_GRH;
4535 grh->flow_label = flow_label;
4536 grh->sgid_index = sgid_index;
4537 grh->hop_limit = hop_limit;
4538 grh->traffic_class = traffic_class;
4539 grh->sgid_attr = NULL;
4542 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4543 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4544 u32 flow_label, u8 hop_limit, u8 traffic_class,
4545 const struct ib_gid_attr *sgid_attr);
4546 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4547 const struct rdma_ah_attr *src);
4548 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4549 const struct rdma_ah_attr *new);
4550 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4553 * rdma_ah_find_type - Return address handle type.
4555 * @dev: Device to be checked
4556 * @port_num: Port number
4558 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4561 if (rdma_protocol_roce(dev, port_num))
4562 return RDMA_AH_ATTR_TYPE_ROCE;
4563 if (rdma_protocol_ib(dev, port_num)) {
4564 if (rdma_cap_opa_ah(dev, port_num))
4565 return RDMA_AH_ATTR_TYPE_OPA;
4566 return RDMA_AH_ATTR_TYPE_IB;
4569 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4573 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4574 * In the current implementation the only way to get
4575 * get the 32bit lid is from other sources for OPA.
4576 * For IB, lids will always be 16bits so cast the
4577 * value accordingly.
4581 static inline u16 ib_lid_cpu16(u32 lid)
4583 WARN_ON_ONCE(lid & 0xFFFF0000);
4588 * ib_lid_be16 - Return lid in 16bit BE encoding.
4592 static inline __be16 ib_lid_be16(u32 lid)
4594 WARN_ON_ONCE(lid & 0xFFFF0000);
4595 return cpu_to_be16((u16)lid);
4599 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4601 * @device: the rdma device
4602 * @comp_vector: index of completion vector
4604 * Returns NULL on failure, otherwise a corresponding cpu map of the
4605 * completion vector (returns all-cpus map if the device driver doesn't
4606 * implement get_vector_affinity).
4608 static inline const struct cpumask *
4609 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4611 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4612 !device->ops.get_vector_affinity)
4615 return device->ops.get_vector_affinity(device, comp_vector);
4620 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4621 * and add their gids, as needed, to the relevant RoCE devices.
4623 * @device: the rdma device
4625 void rdma_roce_rescan_device(struct ib_device *ibdev);
4627 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4629 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4631 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4632 enum rdma_netdev_t type, const char *name,
4633 unsigned char name_assign_type,
4634 void (*setup)(struct net_device *));
4636 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4637 enum rdma_netdev_t type, const char *name,
4638 unsigned char name_assign_type,
4639 void (*setup)(struct net_device *),
4640 struct net_device *netdev);
4643 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4645 * @device: device pointer for which ib_device pointer to retrieve
4647 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4650 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4652 struct ib_core_device *coredev =
4653 container_of(device, struct ib_core_device, dev);
4655 return coredev->owner;
4659 * ibdev_to_node - return the NUMA node for a given ib_device
4660 * @dev: device to get the NUMA node for.
4662 static inline int ibdev_to_node(struct ib_device *ibdev)
4664 struct device *parent = ibdev->dev.parent;
4667 return NUMA_NO_NODE;
4668 return dev_to_node(parent);
4672 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4673 * ib_device holder structure from device pointer.
4675 * NOTE: New drivers should not make use of this API; This API is only for
4676 * existing drivers who have exposed sysfs entries using
4677 * ops->device_group.
4679 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4680 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4682 bool rdma_dev_access_netns(const struct ib_device *device,
4683 const struct net *net);
4685 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4686 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4687 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4690 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4693 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4694 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4697 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4699 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4701 fl_low ^= fl_high >> 14;
4702 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4706 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4707 * local and remote qpn values
4709 * This function folded the multiplication results of two qpns, 24 bit each,
4710 * fields, and converts it to a 20 bit results.
4712 * This function will create symmetric flow_label value based on the local
4713 * and remote qpn values. this will allow both the requester and responder
4714 * to calculate the same flow_label for a given connection.
4716 * This helper function should be used by driver in case the upper layer
4717 * provide a zero flow_label value. This is to improve entropy of RDMA
4718 * traffic in the network.
4720 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4722 u64 v = (u64)lqpn * rqpn;
4727 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4730 const struct ib_port_immutable*
4731 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4732 #endif /* IB_VERBS_H */