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;
2179 struct list_head pkey_list;
2181 struct ib_port_cache cache;
2183 spinlock_t netdev_lock;
2184 struct net_device __rcu *netdev;
2185 struct hlist_node ndev_hash_link;
2186 struct rdma_port_counter port_counter;
2187 struct ib_port *sysfs;
2190 /* rdma netdev type - specifies protocol type */
2191 enum rdma_netdev_t {
2192 RDMA_NETDEV_OPA_VNIC,
2197 * struct rdma_netdev - rdma netdev
2198 * For cases where netstack interfacing is required.
2200 struct rdma_netdev {
2202 struct ib_device *hca;
2207 * cleanup function must be specified.
2208 * FIXME: This is only used for OPA_VNIC and that usage should be
2211 void (*free_rdma_netdev)(struct net_device *netdev);
2213 /* control functions */
2214 void (*set_id)(struct net_device *netdev, int id);
2216 int (*send)(struct net_device *dev, struct sk_buff *skb,
2217 struct ib_ah *address, u32 dqpn);
2219 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2220 union ib_gid *gid, u16 mlid,
2221 int set_qkey, u32 qkey);
2222 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2223 union ib_gid *gid, u16 mlid);
2225 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2228 struct rdma_netdev_alloc_params {
2234 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2235 struct net_device *netdev, void *param);
2238 struct ib_odp_counters {
2240 atomic64_t invalidations;
2241 atomic64_t prefetch;
2244 struct ib_counters {
2245 struct ib_device *device;
2246 struct ib_uobject *uobject;
2247 /* num of objects attached */
2251 struct ib_counters_read_attr {
2254 u32 flags; /* use enum ib_read_counters_flags */
2257 struct uverbs_attr_bundle;
2259 struct iw_cm_conn_param;
2261 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2262 .size_##ib_struct = \
2263 (sizeof(struct drv_struct) + \
2264 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2265 BUILD_BUG_ON_ZERO( \
2266 !__same_type(((struct drv_struct *)NULL)->member, \
2269 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2270 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2272 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2273 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2275 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2277 struct rdma_user_mmap_entry {
2279 struct ib_ucontext *ucontext;
2280 unsigned long start_pgoff;
2282 bool driver_removed;
2285 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2287 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2289 return (u64)entry->start_pgoff << PAGE_SHIFT;
2293 * struct ib_device_ops - InfiniBand device operations
2294 * This structure defines all the InfiniBand device operations, providers will
2295 * need to define the supported operations, otherwise they will be set to null.
2297 struct ib_device_ops {
2298 struct module *owner;
2299 enum rdma_driver_id driver_id;
2301 unsigned int uverbs_no_driver_id_binding:1;
2303 const struct attribute_group **port_groups;
2305 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2306 const struct ib_send_wr **bad_send_wr);
2307 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2308 const struct ib_recv_wr **bad_recv_wr);
2309 void (*drain_rq)(struct ib_qp *qp);
2310 void (*drain_sq)(struct ib_qp *qp);
2311 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2312 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2313 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2314 int (*post_srq_recv)(struct ib_srq *srq,
2315 const struct ib_recv_wr *recv_wr,
2316 const struct ib_recv_wr **bad_recv_wr);
2317 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2318 u32 port_num, const struct ib_wc *in_wc,
2319 const struct ib_grh *in_grh,
2320 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2321 size_t *out_mad_size, u16 *out_mad_pkey_index);
2322 int (*query_device)(struct ib_device *device,
2323 struct ib_device_attr *device_attr,
2324 struct ib_udata *udata);
2325 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2326 struct ib_device_modify *device_modify);
2327 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2328 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2330 int (*query_port)(struct ib_device *device, u32 port_num,
2331 struct ib_port_attr *port_attr);
2332 int (*modify_port)(struct ib_device *device, u32 port_num,
2333 int port_modify_mask,
2334 struct ib_port_modify *port_modify);
2336 * The following mandatory functions are used only at device
2337 * registration. Keep functions such as these at the end of this
2338 * structure to avoid cache line misses when accessing struct ib_device
2341 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2342 struct ib_port_immutable *immutable);
2343 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2346 * When calling get_netdev, the HW vendor's driver should return the
2347 * net device of device @device at port @port_num or NULL if such
2348 * a net device doesn't exist. The vendor driver should call dev_hold
2349 * on this net device. The HW vendor's device driver must guarantee
2350 * that this function returns NULL before the net device has finished
2351 * NETDEV_UNREGISTER state.
2353 struct net_device *(*get_netdev)(struct ib_device *device,
2356 * rdma netdev operation
2358 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2359 * must return -EOPNOTSUPP if it doesn't support the specified type.
2361 struct net_device *(*alloc_rdma_netdev)(
2362 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2363 const char *name, unsigned char name_assign_type,
2364 void (*setup)(struct net_device *));
2366 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2367 enum rdma_netdev_t type,
2368 struct rdma_netdev_alloc_params *params);
2370 * query_gid should be return GID value for @device, when @port_num
2371 * link layer is either IB or iWarp. It is no-op if @port_num port
2372 * is RoCE link layer.
2374 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2377 * When calling add_gid, the HW vendor's driver should add the gid
2378 * of device of port at gid index available at @attr. Meta-info of
2379 * that gid (for example, the network device related to this gid) is
2380 * available at @attr. @context allows the HW vendor driver to store
2381 * extra information together with a GID entry. The HW vendor driver may
2382 * allocate memory to contain this information and store it in @context
2383 * when a new GID entry is written to. Params are consistent until the
2384 * next call of add_gid or delete_gid. The function should return 0 on
2385 * success or error otherwise. The function could be called
2386 * concurrently for different ports. This function is only called when
2387 * roce_gid_table is used.
2389 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2391 * When calling del_gid, the HW vendor's driver should delete the
2392 * gid of device @device at gid index gid_index of port port_num
2393 * available in @attr.
2394 * Upon the deletion of a GID entry, the HW vendor must free any
2395 * allocated memory. The caller will clear @context afterwards.
2396 * This function is only called when roce_gid_table is used.
2398 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2399 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2401 int (*alloc_ucontext)(struct ib_ucontext *context,
2402 struct ib_udata *udata);
2403 void (*dealloc_ucontext)(struct ib_ucontext *context);
2404 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2406 * This will be called once refcount of an entry in mmap_xa reaches
2407 * zero. The type of the memory that was mapped may differ between
2408 * entries and is opaque to the rdma_user_mmap interface.
2409 * Therefore needs to be implemented by the driver in mmap_free.
2411 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2412 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2413 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2414 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2415 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2416 struct ib_udata *udata);
2417 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2418 struct ib_udata *udata);
2419 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2420 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2421 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2422 int (*create_srq)(struct ib_srq *srq,
2423 struct ib_srq_init_attr *srq_init_attr,
2424 struct ib_udata *udata);
2425 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2426 enum ib_srq_attr_mask srq_attr_mask,
2427 struct ib_udata *udata);
2428 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2429 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2430 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2431 struct ib_qp_init_attr *qp_init_attr,
2432 struct ib_udata *udata);
2433 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2434 int qp_attr_mask, struct ib_udata *udata);
2435 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2436 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2437 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2438 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2439 struct ib_udata *udata);
2440 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2441 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2442 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2443 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2444 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2445 u64 virt_addr, int mr_access_flags,
2446 struct ib_udata *udata);
2447 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2448 u64 length, u64 virt_addr, int fd,
2449 int mr_access_flags,
2450 struct ib_udata *udata);
2451 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2452 u64 length, u64 virt_addr,
2453 int mr_access_flags, struct ib_pd *pd,
2454 struct ib_udata *udata);
2455 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2456 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2458 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2459 u32 max_num_data_sg,
2460 u32 max_num_meta_sg);
2461 int (*advise_mr)(struct ib_pd *pd,
2462 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2463 struct ib_sge *sg_list, u32 num_sge,
2464 struct uverbs_attr_bundle *attrs);
2465 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2466 unsigned int *sg_offset);
2467 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2468 struct ib_mr_status *mr_status);
2469 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2470 int (*dealloc_mw)(struct ib_mw *mw);
2471 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2472 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2473 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2474 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2475 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2476 struct ib_flow_attr *flow_attr,
2477 struct ib_udata *udata);
2478 int (*destroy_flow)(struct ib_flow *flow_id);
2479 struct ib_flow_action *(*create_flow_action_esp)(
2480 struct ib_device *device,
2481 const struct ib_flow_action_attrs_esp *attr,
2482 struct uverbs_attr_bundle *attrs);
2483 int (*destroy_flow_action)(struct ib_flow_action *action);
2484 int (*modify_flow_action_esp)(
2485 struct ib_flow_action *action,
2486 const struct ib_flow_action_attrs_esp *attr,
2487 struct uverbs_attr_bundle *attrs);
2488 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2490 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2491 struct ifla_vf_info *ivf);
2492 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2493 struct ifla_vf_stats *stats);
2494 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2495 struct ifla_vf_guid *node_guid,
2496 struct ifla_vf_guid *port_guid);
2497 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2499 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2500 struct ib_wq_init_attr *init_attr,
2501 struct ib_udata *udata);
2502 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2503 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2504 u32 wq_attr_mask, struct ib_udata *udata);
2505 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2506 struct ib_rwq_ind_table_init_attr *init_attr,
2507 struct ib_udata *udata);
2508 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2509 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2510 struct ib_ucontext *context,
2511 struct ib_dm_alloc_attr *attr,
2512 struct uverbs_attr_bundle *attrs);
2513 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2514 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2515 struct ib_dm_mr_attr *attr,
2516 struct uverbs_attr_bundle *attrs);
2517 int (*create_counters)(struct ib_counters *counters,
2518 struct uverbs_attr_bundle *attrs);
2519 int (*destroy_counters)(struct ib_counters *counters);
2520 int (*read_counters)(struct ib_counters *counters,
2521 struct ib_counters_read_attr *counters_read_attr,
2522 struct uverbs_attr_bundle *attrs);
2523 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2524 int data_sg_nents, unsigned int *data_sg_offset,
2525 struct scatterlist *meta_sg, int meta_sg_nents,
2526 unsigned int *meta_sg_offset);
2529 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2530 * fill in the driver initialized data. The struct is kfree()'ed by
2531 * the sysfs core when the device is removed. A lifespan of -1 in the
2532 * return struct tells the core to set a default lifespan.
2534 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2535 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2538 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2539 * @index - The index in the value array we wish to have updated, or
2540 * num_counters if we want all stats updated
2542 * < 0 - Error, no counters updated
2543 * index - Updated the single counter pointed to by index
2544 * num_counters - Updated all counters (will reset the timestamp
2545 * and prevent further calls for lifespan milliseconds)
2546 * Drivers are allowed to update all counters in leiu of just the
2547 * one given in index at their option
2549 int (*get_hw_stats)(struct ib_device *device,
2550 struct rdma_hw_stats *stats, u32 port, int index);
2553 * Allows rdma drivers to add their own restrack attributes.
2555 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2556 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2557 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2558 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2559 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2560 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2561 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2563 /* Device lifecycle callbacks */
2565 * Called after the device becomes registered, before clients are
2568 int (*enable_driver)(struct ib_device *dev);
2570 * This is called as part of ib_dealloc_device().
2572 void (*dealloc_driver)(struct ib_device *dev);
2574 /* iWarp CM callbacks */
2575 void (*iw_add_ref)(struct ib_qp *qp);
2576 void (*iw_rem_ref)(struct ib_qp *qp);
2577 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2578 int (*iw_connect)(struct iw_cm_id *cm_id,
2579 struct iw_cm_conn_param *conn_param);
2580 int (*iw_accept)(struct iw_cm_id *cm_id,
2581 struct iw_cm_conn_param *conn_param);
2582 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2584 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2585 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2587 * counter_bind_qp - Bind a QP to a counter.
2588 * @counter - The counter to be bound. If counter->id is zero then
2589 * the driver needs to allocate a new counter and set counter->id
2591 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2593 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2594 * counter and bind it onto the default one
2596 int (*counter_unbind_qp)(struct ib_qp *qp);
2598 * counter_dealloc -De-allocate the hw counter
2600 int (*counter_dealloc)(struct rdma_counter *counter);
2602 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2603 * the driver initialized data.
2605 struct rdma_hw_stats *(*counter_alloc_stats)(
2606 struct rdma_counter *counter);
2608 * counter_update_stats - Query the stats value of this counter
2610 int (*counter_update_stats)(struct rdma_counter *counter);
2613 * Allows rdma drivers to add their own restrack attributes
2614 * dumped via 'rdma stat' iproute2 command.
2616 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2618 /* query driver for its ucontext properties */
2619 int (*query_ucontext)(struct ib_ucontext *context,
2620 struct uverbs_attr_bundle *attrs);
2622 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2623 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2624 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2625 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2626 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2627 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2628 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2629 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2630 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2633 struct ib_core_device {
2634 /* device must be the first element in structure until,
2635 * union of ib_core_device and device exists in ib_device.
2638 possible_net_t rdma_net;
2639 struct kobject *ports_kobj;
2640 struct list_head port_list;
2641 struct ib_device *owner; /* reach back to owner ib_device */
2644 struct rdma_restrack_root;
2646 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2647 struct device *dma_device;
2648 struct ib_device_ops ops;
2649 char name[IB_DEVICE_NAME_MAX];
2650 struct rcu_head rcu_head;
2652 struct list_head event_handler_list;
2653 /* Protects event_handler_list */
2654 struct rw_semaphore event_handler_rwsem;
2656 /* Protects QP's event_handler calls and open_qp list */
2657 spinlock_t qp_open_list_lock;
2659 struct rw_semaphore client_data_rwsem;
2660 struct xarray client_data;
2661 struct mutex unregistration_lock;
2663 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2664 rwlock_t cache_lock;
2666 * port_data is indexed by port number
2668 struct ib_port_data *port_data;
2670 int num_comp_vectors;
2674 struct ib_core_device coredev;
2677 /* First group is for device attributes,
2678 * Second group is for driver provided attributes (optional).
2679 * Third group is for the hw_stats
2680 * It is a NULL terminated array.
2682 const struct attribute_group *groups[4];
2684 u64 uverbs_cmd_mask;
2686 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2690 /* Indicates kernel verbs support, should not be used in drivers */
2691 u16 kverbs_provider:1;
2692 /* CQ adaptive moderation (RDMA DIM) */
2696 struct ib_device_attr attrs;
2697 struct hw_stats_device_data *hw_stats_data;
2699 #ifdef CONFIG_CGROUP_RDMA
2700 struct rdmacg_device cg_device;
2705 spinlock_t cq_pools_lock;
2706 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2708 struct rdma_restrack_root *res;
2710 const struct uapi_definition *driver_def;
2713 * Positive refcount indicates that the device is currently
2714 * registered and cannot be unregistered.
2716 refcount_t refcount;
2717 struct completion unreg_completion;
2718 struct work_struct unregistration_work;
2720 const struct rdma_link_ops *link_ops;
2722 /* Protects compat_devs xarray modifications */
2723 struct mutex compat_devs_mutex;
2724 /* Maintains compat devices for each net namespace */
2725 struct xarray compat_devs;
2727 /* Used by iWarp CM */
2728 char iw_ifname[IFNAMSIZ];
2729 u32 iw_driver_flags;
2733 struct ib_client_nl_info;
2736 int (*add)(struct ib_device *ibdev);
2737 void (*remove)(struct ib_device *, void *client_data);
2738 void (*rename)(struct ib_device *dev, void *client_data);
2739 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2740 struct ib_client_nl_info *res);
2741 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2743 /* Returns the net_dev belonging to this ib_client and matching the
2745 * @dev: An RDMA device that the net_dev use for communication.
2746 * @port: A physical port number on the RDMA device.
2747 * @pkey: P_Key that the net_dev uses if applicable.
2748 * @gid: A GID that the net_dev uses to communicate.
2749 * @addr: An IP address the net_dev is configured with.
2750 * @client_data: The device's client data set by ib_set_client_data().
2752 * An ib_client that implements a net_dev on top of RDMA devices
2753 * (such as IP over IB) should implement this callback, allowing the
2754 * rdma_cm module to find the right net_dev for a given request.
2756 * The caller is responsible for calling dev_put on the returned
2758 struct net_device *(*get_net_dev_by_params)(
2759 struct ib_device *dev,
2762 const union ib_gid *gid,
2763 const struct sockaddr *addr,
2767 struct completion uses_zero;
2770 /* kverbs are not required by the client */
2775 * IB block DMA iterator
2777 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2778 * to a HW supported page size.
2780 struct ib_block_iter {
2781 /* internal states */
2782 struct scatterlist *__sg; /* sg holding the current aligned block */
2783 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2784 unsigned int __sg_nents; /* number of SG entries */
2785 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2786 unsigned int __pg_bit; /* alignment of current block */
2789 struct ib_device *_ib_alloc_device(size_t size);
2790 #define ib_alloc_device(drv_struct, member) \
2791 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2792 BUILD_BUG_ON_ZERO(offsetof( \
2793 struct drv_struct, member))), \
2794 struct drv_struct, member)
2796 void ib_dealloc_device(struct ib_device *device);
2798 void ib_get_device_fw_str(struct ib_device *device, char *str);
2800 int ib_register_device(struct ib_device *device, const char *name,
2801 struct device *dma_device);
2802 void ib_unregister_device(struct ib_device *device);
2803 void ib_unregister_driver(enum rdma_driver_id driver_id);
2804 void ib_unregister_device_and_put(struct ib_device *device);
2805 void ib_unregister_device_queued(struct ib_device *ib_dev);
2807 int ib_register_client (struct ib_client *client);
2808 void ib_unregister_client(struct ib_client *client);
2810 void __rdma_block_iter_start(struct ib_block_iter *biter,
2811 struct scatterlist *sglist,
2813 unsigned long pgsz);
2814 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2817 * rdma_block_iter_dma_address - get the aligned dma address of the current
2818 * block held by the block iterator.
2819 * @biter: block iterator holding the memory block
2821 static inline dma_addr_t
2822 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2824 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2828 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2829 * @sglist: sglist to iterate over
2830 * @biter: block iterator holding the memory block
2831 * @nents: maximum number of sg entries to iterate over
2832 * @pgsz: best HW supported page size to use
2834 * Callers may use rdma_block_iter_dma_address() to get each
2835 * blocks aligned DMA address.
2837 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2838 for (__rdma_block_iter_start(biter, sglist, nents, \
2840 __rdma_block_iter_next(biter);)
2843 * ib_get_client_data - Get IB client context
2844 * @device:Device to get context for
2845 * @client:Client to get context for
2847 * ib_get_client_data() returns the client context data set with
2848 * ib_set_client_data(). This can only be called while the client is
2849 * registered to the device, once the ib_client remove() callback returns this
2852 static inline void *ib_get_client_data(struct ib_device *device,
2853 struct ib_client *client)
2855 return xa_load(&device->client_data, client->client_id);
2857 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2859 void ib_set_device_ops(struct ib_device *device,
2860 const struct ib_device_ops *ops);
2862 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2863 unsigned long pfn, unsigned long size, pgprot_t prot,
2864 struct rdma_user_mmap_entry *entry);
2865 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2866 struct rdma_user_mmap_entry *entry,
2868 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2869 struct rdma_user_mmap_entry *entry,
2870 size_t length, u32 min_pgoff,
2873 struct rdma_user_mmap_entry *
2874 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2875 unsigned long pgoff);
2876 struct rdma_user_mmap_entry *
2877 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2878 struct vm_area_struct *vma);
2879 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2881 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2883 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2885 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2888 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2890 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2893 static inline bool ib_is_buffer_cleared(const void __user *p,
2899 if (len > USHRT_MAX)
2902 buf = memdup_user(p, len);
2906 ret = !memchr_inv(buf, 0, len);
2911 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2915 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2919 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2920 * contains all required attributes and no attributes not allowed for
2921 * the given QP state transition.
2922 * @cur_state: Current QP state
2923 * @next_state: Next QP state
2925 * @mask: Mask of supplied QP attributes
2927 * This function is a helper function that a low-level driver's
2928 * modify_qp method can use to validate the consumer's input. It
2929 * checks that cur_state and next_state are valid QP states, that a
2930 * transition from cur_state to next_state is allowed by the IB spec,
2931 * and that the attribute mask supplied is allowed for the transition.
2933 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2934 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2936 void ib_register_event_handler(struct ib_event_handler *event_handler);
2937 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2938 void ib_dispatch_event(const struct ib_event *event);
2940 int ib_query_port(struct ib_device *device,
2941 u32 port_num, struct ib_port_attr *port_attr);
2943 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2947 * rdma_cap_ib_switch - Check if the device is IB switch
2948 * @device: Device to check
2950 * Device driver is responsible for setting is_switch bit on
2951 * in ib_device structure at init time.
2953 * Return: true if the device is IB switch.
2955 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2957 return device->is_switch;
2961 * rdma_start_port - Return the first valid port number for the device
2964 * @device: Device to be checked
2966 * Return start port number
2968 static inline u32 rdma_start_port(const struct ib_device *device)
2970 return rdma_cap_ib_switch(device) ? 0 : 1;
2974 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2975 * @device - The struct ib_device * to iterate over
2976 * @iter - The unsigned int to store the port number
2978 #define rdma_for_each_port(device, iter) \
2979 for (iter = rdma_start_port(device + \
2980 BUILD_BUG_ON_ZERO(!__same_type(u32, \
2982 iter <= rdma_end_port(device); iter++)
2985 * rdma_end_port - Return the last valid port number for the device
2988 * @device: Device to be checked
2990 * Return last port number
2992 static inline u32 rdma_end_port(const struct ib_device *device)
2994 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2997 static inline int rdma_is_port_valid(const struct ib_device *device,
3000 return (port >= rdma_start_port(device) &&
3001 port <= rdma_end_port(device));
3004 static inline bool rdma_is_grh_required(const struct ib_device *device,
3007 return device->port_data[port_num].immutable.core_cap_flags &
3008 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3011 static inline bool rdma_protocol_ib(const struct ib_device *device,
3014 return device->port_data[port_num].immutable.core_cap_flags &
3015 RDMA_CORE_CAP_PROT_IB;
3018 static inline bool rdma_protocol_roce(const struct ib_device *device,
3021 return device->port_data[port_num].immutable.core_cap_flags &
3022 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3025 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3028 return device->port_data[port_num].immutable.core_cap_flags &
3029 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3032 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3035 return device->port_data[port_num].immutable.core_cap_flags &
3036 RDMA_CORE_CAP_PROT_ROCE;
3039 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3042 return device->port_data[port_num].immutable.core_cap_flags &
3043 RDMA_CORE_CAP_PROT_IWARP;
3046 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3049 return rdma_protocol_ib(device, port_num) ||
3050 rdma_protocol_roce(device, port_num);
3053 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3056 return device->port_data[port_num].immutable.core_cap_flags &
3057 RDMA_CORE_CAP_PROT_RAW_PACKET;
3060 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3063 return device->port_data[port_num].immutable.core_cap_flags &
3064 RDMA_CORE_CAP_PROT_USNIC;
3068 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3069 * Management Datagrams.
3070 * @device: Device to check
3071 * @port_num: Port number to check
3073 * Management Datagrams (MAD) are a required part of the InfiniBand
3074 * specification and are supported on all InfiniBand devices. A slightly
3075 * extended version are also supported on OPA interfaces.
3077 * Return: true if the port supports sending/receiving of MAD packets.
3079 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3081 return device->port_data[port_num].immutable.core_cap_flags &
3082 RDMA_CORE_CAP_IB_MAD;
3086 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3087 * Management Datagrams.
3088 * @device: Device to check
3089 * @port_num: Port number to check
3091 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3092 * datagrams with their own versions. These OPA MADs share many but not all of
3093 * the characteristics of InfiniBand MADs.
3095 * OPA MADs differ in the following ways:
3097 * 1) MADs are variable size up to 2K
3098 * IBTA defined MADs remain fixed at 256 bytes
3099 * 2) OPA SMPs must carry valid PKeys
3100 * 3) OPA SMP packets are a different format
3102 * Return: true if the port supports OPA MAD packet formats.
3104 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3106 return device->port_data[port_num].immutable.core_cap_flags &
3107 RDMA_CORE_CAP_OPA_MAD;
3111 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3112 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3113 * @device: Device to check
3114 * @port_num: Port number to check
3116 * Each InfiniBand node is required to provide a Subnet Management Agent
3117 * that the subnet manager can access. Prior to the fabric being fully
3118 * configured by the subnet manager, the SMA is accessed via a well known
3119 * interface called the Subnet Management Interface (SMI). This interface
3120 * uses directed route packets to communicate with the SM to get around the
3121 * chicken and egg problem of the SM needing to know what's on the fabric
3122 * in order to configure the fabric, and needing to configure the fabric in
3123 * order to send packets to the devices on the fabric. These directed
3124 * route packets do not need the fabric fully configured in order to reach
3125 * their destination. The SMI is the only method allowed to send
3126 * directed route packets on an InfiniBand fabric.
3128 * Return: true if the port provides an SMI.
3130 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3132 return device->port_data[port_num].immutable.core_cap_flags &
3133 RDMA_CORE_CAP_IB_SMI;
3137 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3138 * Communication Manager.
3139 * @device: Device to check
3140 * @port_num: Port number to check
3142 * The InfiniBand Communication Manager is one of many pre-defined General
3143 * Service Agents (GSA) that are accessed via the General Service
3144 * Interface (GSI). It's role is to facilitate establishment of connections
3145 * between nodes as well as other management related tasks for established
3148 * Return: true if the port supports an IB CM (this does not guarantee that
3149 * a CM is actually running however).
3151 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3153 return device->port_data[port_num].immutable.core_cap_flags &
3154 RDMA_CORE_CAP_IB_CM;
3158 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3159 * Communication Manager.
3160 * @device: Device to check
3161 * @port_num: Port number to check
3163 * Similar to above, but specific to iWARP connections which have a different
3164 * managment protocol than InfiniBand.
3166 * Return: true if the port supports an iWARP CM (this does not guarantee that
3167 * a CM is actually running however).
3169 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3171 return device->port_data[port_num].immutable.core_cap_flags &
3172 RDMA_CORE_CAP_IW_CM;
3176 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3177 * Subnet Administration.
3178 * @device: Device to check
3179 * @port_num: Port number to check
3181 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3182 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3183 * fabrics, devices should resolve routes to other hosts by contacting the
3184 * SA to query the proper route.
3186 * Return: true if the port should act as a client to the fabric Subnet
3187 * Administration interface. This does not imply that the SA service is
3190 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3192 return device->port_data[port_num].immutable.core_cap_flags &
3193 RDMA_CORE_CAP_IB_SA;
3197 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3199 * @device: Device to check
3200 * @port_num: Port number to check
3202 * InfiniBand multicast registration is more complex than normal IPv4 or
3203 * IPv6 multicast registration. Each Host Channel Adapter must register
3204 * with the Subnet Manager when it wishes to join a multicast group. It
3205 * should do so only once regardless of how many queue pairs it subscribes
3206 * to this group. And it should leave the group only after all queue pairs
3207 * attached to the group have been detached.
3209 * Return: true if the port must undertake the additional adminstrative
3210 * overhead of registering/unregistering with the SM and tracking of the
3211 * total number of queue pairs attached to the multicast group.
3213 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3216 return rdma_cap_ib_sa(device, port_num);
3220 * rdma_cap_af_ib - Check if the port of device has the capability
3221 * Native Infiniband Address.
3222 * @device: Device to check
3223 * @port_num: Port number to check
3225 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3226 * GID. RoCE uses a different mechanism, but still generates a GID via
3227 * a prescribed mechanism and port specific data.
3229 * Return: true if the port uses a GID address to identify devices on the
3232 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3234 return device->port_data[port_num].immutable.core_cap_flags &
3235 RDMA_CORE_CAP_AF_IB;
3239 * rdma_cap_eth_ah - Check if the port of device has the capability
3240 * Ethernet Address Handle.
3241 * @device: Device to check
3242 * @port_num: Port number to check
3244 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3245 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3246 * port. Normally, packet headers are generated by the sending host
3247 * adapter, but when sending connectionless datagrams, we must manually
3248 * inject the proper headers for the fabric we are communicating over.
3250 * Return: true if we are running as a RoCE port and must force the
3251 * addition of a Global Route Header built from our Ethernet Address
3252 * Handle into our header list for connectionless packets.
3254 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3256 return device->port_data[port_num].immutable.core_cap_flags &
3257 RDMA_CORE_CAP_ETH_AH;
3261 * rdma_cap_opa_ah - Check if the port of device supports
3262 * OPA Address handles
3263 * @device: Device to check
3264 * @port_num: Port number to check
3266 * Return: true if we are running on an OPA device which supports
3267 * the extended OPA addressing.
3269 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3271 return (device->port_data[port_num].immutable.core_cap_flags &
3272 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3276 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3279 * @port_num: Port number
3281 * This MAD size includes the MAD headers and MAD payload. No other headers
3284 * Return the max MAD size required by the Port. Will return 0 if the port
3285 * does not support MADs
3287 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3290 return device->port_data[port_num].immutable.max_mad_size;
3294 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3295 * @device: Device to check
3296 * @port_num: Port number to check
3298 * RoCE GID table mechanism manages the various GIDs for a device.
3300 * NOTE: if allocating the port's GID table has failed, this call will still
3301 * return true, but any RoCE GID table API will fail.
3303 * Return: true if the port uses RoCE GID table mechanism in order to manage
3306 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3309 return rdma_protocol_roce(device, port_num) &&
3310 device->ops.add_gid && device->ops.del_gid;
3314 * Check if the device supports READ W/ INVALIDATE.
3316 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3319 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3320 * has support for it yet.
3322 return rdma_protocol_iwarp(dev, port_num);
3326 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3328 * @port_num: 1 based Port number
3330 * Return true if port is an Intel OPA port , false if not
3332 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3335 return (device->port_data[port_num].immutable.core_cap_flags &
3336 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3340 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3342 * @port_num: Port number
3343 * @mtu: enum value of MTU
3345 * Return the MTU size supported by the port as an integer value. Will return
3346 * -1 if enum value of mtu is not supported.
3348 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3351 if (rdma_core_cap_opa_port(device, port))
3352 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3354 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3358 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3360 * @port_num: Port number
3361 * @attr: port attribute
3363 * Return the MTU size supported by the port as an integer value.
3365 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3366 struct ib_port_attr *attr)
3368 if (rdma_core_cap_opa_port(device, port))
3369 return attr->phys_mtu;
3371 return ib_mtu_enum_to_int(attr->max_mtu);
3374 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3376 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3377 struct ifla_vf_info *info);
3378 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3379 struct ifla_vf_stats *stats);
3380 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3381 struct ifla_vf_guid *node_guid,
3382 struct ifla_vf_guid *port_guid);
3383 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3386 int ib_query_pkey(struct ib_device *device,
3387 u32 port_num, u16 index, u16 *pkey);
3389 int ib_modify_device(struct ib_device *device,
3390 int device_modify_mask,
3391 struct ib_device_modify *device_modify);
3393 int ib_modify_port(struct ib_device *device,
3394 u32 port_num, int port_modify_mask,
3395 struct ib_port_modify *port_modify);
3397 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3398 u32 *port_num, u16 *index);
3400 int ib_find_pkey(struct ib_device *device,
3401 u32 port_num, u16 pkey, u16 *index);
3405 * Create a memory registration for all memory in the system and place
3406 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3407 * ULPs to avoid the overhead of dynamic MRs.
3409 * This flag is generally considered unsafe and must only be used in
3410 * extremly trusted environments. Every use of it will log a warning
3411 * in the kernel log.
3413 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3416 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3417 const char *caller);
3420 * ib_alloc_pd - Allocates an unused protection domain.
3421 * @device: The device on which to allocate the protection domain.
3422 * @flags: protection domain flags
3424 * A protection domain object provides an association between QPs, shared
3425 * receive queues, address handles, memory regions, and memory windows.
3427 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3428 * memory operations.
3430 #define ib_alloc_pd(device, flags) \
3431 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3433 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3436 * ib_dealloc_pd - Deallocate kernel PD
3437 * @pd: The protection domain
3439 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3441 static inline void ib_dealloc_pd(struct ib_pd *pd)
3443 int ret = ib_dealloc_pd_user(pd, NULL);
3445 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3448 enum rdma_create_ah_flags {
3449 /* In a sleepable context */
3450 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3454 * rdma_create_ah - Creates an address handle for the given address vector.
3455 * @pd: The protection domain associated with the address handle.
3456 * @ah_attr: The attributes of the address vector.
3457 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3459 * The address handle is used to reference a local or global destination
3460 * in all UD QP post sends.
3462 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3466 * rdma_create_user_ah - Creates an address handle for the given address vector.
3467 * It resolves destination mac address for ah attribute of RoCE type.
3468 * @pd: The protection domain associated with the address handle.
3469 * @ah_attr: The attributes of the address vector.
3470 * @udata: pointer to user's input output buffer information need by
3473 * It returns 0 on success and returns appropriate error code on error.
3474 * The address handle is used to reference a local or global destination
3475 * in all UD QP post sends.
3477 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3478 struct rdma_ah_attr *ah_attr,
3479 struct ib_udata *udata);
3481 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3483 * @hdr: the L3 header to parse
3484 * @net_type: type of header to parse
3485 * @sgid: place to store source gid
3486 * @dgid: place to store destination gid
3488 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3489 enum rdma_network_type net_type,
3490 union ib_gid *sgid, union ib_gid *dgid);
3493 * ib_get_rdma_header_version - Get the header version
3494 * @hdr: the L3 header to parse
3496 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3499 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3501 * @device: Device on which the received message arrived.
3502 * @port_num: Port on which the received message arrived.
3503 * @wc: Work completion associated with the received message.
3504 * @grh: References the received global route header. This parameter is
3505 * ignored unless the work completion indicates that the GRH is valid.
3506 * @ah_attr: Returned attributes that can be used when creating an address
3507 * handle for replying to the message.
3508 * When ib_init_ah_attr_from_wc() returns success,
3509 * (a) for IB link layer it optionally contains a reference to SGID attribute
3510 * when GRH is present for IB link layer.
3511 * (b) for RoCE link layer it contains a reference to SGID attribute.
3512 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3513 * attributes which are initialized using ib_init_ah_attr_from_wc().
3516 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3517 const struct ib_wc *wc, const struct ib_grh *grh,
3518 struct rdma_ah_attr *ah_attr);
3521 * ib_create_ah_from_wc - Creates an address handle associated with the
3522 * sender of the specified work completion.
3523 * @pd: The protection domain associated with the address handle.
3524 * @wc: Work completion information associated with a received message.
3525 * @grh: References the received global route header. This parameter is
3526 * ignored unless the work completion indicates that the GRH is valid.
3527 * @port_num: The outbound port number to associate with the address.
3529 * The address handle is used to reference a local or global destination
3530 * in all UD QP post sends.
3532 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3533 const struct ib_grh *grh, u32 port_num);
3536 * rdma_modify_ah - Modifies the address vector associated with an address
3538 * @ah: The address handle to modify.
3539 * @ah_attr: The new address vector attributes to associate with the
3542 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3545 * rdma_query_ah - Queries the address vector associated with an address
3547 * @ah: The address handle to query.
3548 * @ah_attr: The address vector attributes associated with the address
3551 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3553 enum rdma_destroy_ah_flags {
3554 /* In a sleepable context */
3555 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3559 * rdma_destroy_ah_user - Destroys an address handle.
3560 * @ah: The address handle to destroy.
3561 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3562 * @udata: Valid user data or NULL for kernel objects
3564 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3567 * rdma_destroy_ah - Destroys an kernel address handle.
3568 * @ah: The address handle to destroy.
3569 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3571 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3573 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3575 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3577 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3580 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3581 struct ib_srq_init_attr *srq_init_attr,
3582 struct ib_usrq_object *uobject,
3583 struct ib_udata *udata);
3584 static inline struct ib_srq *
3585 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3587 if (!pd->device->ops.create_srq)
3588 return ERR_PTR(-EOPNOTSUPP);
3590 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3594 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3595 * @srq: The SRQ to modify.
3596 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3597 * the current values of selected SRQ attributes are returned.
3598 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3599 * are being modified.
3601 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3602 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3603 * the number of receives queued drops below the limit.
3605 int ib_modify_srq(struct ib_srq *srq,
3606 struct ib_srq_attr *srq_attr,
3607 enum ib_srq_attr_mask srq_attr_mask);
3610 * ib_query_srq - Returns the attribute list and current values for the
3612 * @srq: The SRQ to query.
3613 * @srq_attr: The attributes of the specified SRQ.
3615 int ib_query_srq(struct ib_srq *srq,
3616 struct ib_srq_attr *srq_attr);
3619 * ib_destroy_srq_user - Destroys the specified SRQ.
3620 * @srq: The SRQ to destroy.
3621 * @udata: Valid user data or NULL for kernel objects
3623 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3626 * ib_destroy_srq - Destroys the specified kernel SRQ.
3627 * @srq: The SRQ to destroy.
3629 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3631 static inline void ib_destroy_srq(struct ib_srq *srq)
3633 int ret = ib_destroy_srq_user(srq, NULL);
3635 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3639 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3640 * @srq: The SRQ to post the work request on.
3641 * @recv_wr: A list of work requests to post on the receive queue.
3642 * @bad_recv_wr: On an immediate failure, this parameter will reference
3643 * the work request that failed to be posted on the QP.
3645 static inline int ib_post_srq_recv(struct ib_srq *srq,
3646 const struct ib_recv_wr *recv_wr,
3647 const struct ib_recv_wr **bad_recv_wr)
3649 const struct ib_recv_wr *dummy;
3651 return srq->device->ops.post_srq_recv(srq, recv_wr,
3652 bad_recv_wr ? : &dummy);
3655 struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
3656 struct ib_qp_init_attr *qp_init_attr,
3657 const char *caller);
3658 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3659 struct ib_qp_init_attr *init_attr)
3661 return ib_create_named_qp(pd, init_attr, KBUILD_MODNAME);
3665 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3666 * @qp: The QP to modify.
3667 * @attr: On input, specifies the QP attributes to modify. On output,
3668 * the current values of selected QP attributes are returned.
3669 * @attr_mask: A bit-mask used to specify which attributes of the QP
3670 * are being modified.
3671 * @udata: pointer to user's input output buffer information
3672 * are being modified.
3673 * It returns 0 on success and returns appropriate error code on error.
3675 int ib_modify_qp_with_udata(struct ib_qp *qp,
3676 struct ib_qp_attr *attr,
3678 struct ib_udata *udata);
3681 * ib_modify_qp - Modifies the attributes for the specified QP and then
3682 * transitions the QP to the given state.
3683 * @qp: The QP to modify.
3684 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3685 * the current values of selected QP attributes are returned.
3686 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3687 * are being modified.
3689 int ib_modify_qp(struct ib_qp *qp,
3690 struct ib_qp_attr *qp_attr,
3694 * ib_query_qp - Returns the attribute list and current values for the
3696 * @qp: The QP to query.
3697 * @qp_attr: The attributes of the specified QP.
3698 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3699 * @qp_init_attr: Additional attributes of the selected QP.
3701 * The qp_attr_mask may be used to limit the query to gathering only the
3702 * selected attributes.
3704 int ib_query_qp(struct ib_qp *qp,
3705 struct ib_qp_attr *qp_attr,
3707 struct ib_qp_init_attr *qp_init_attr);
3710 * ib_destroy_qp - Destroys the specified QP.
3711 * @qp: The QP to destroy.
3712 * @udata: Valid udata or NULL for kernel objects
3714 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3717 * ib_destroy_qp - Destroys the specified kernel QP.
3718 * @qp: The QP to destroy.
3720 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3722 static inline int ib_destroy_qp(struct ib_qp *qp)
3724 return ib_destroy_qp_user(qp, NULL);
3728 * ib_open_qp - Obtain a reference to an existing sharable QP.
3729 * @xrcd - XRC domain
3730 * @qp_open_attr: Attributes identifying the QP to open.
3732 * Returns a reference to a sharable QP.
3734 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3735 struct ib_qp_open_attr *qp_open_attr);
3738 * ib_close_qp - Release an external reference to a QP.
3739 * @qp: The QP handle to release
3741 * The opened QP handle is released by the caller. The underlying
3742 * shared QP is not destroyed until all internal references are released.
3744 int ib_close_qp(struct ib_qp *qp);
3747 * ib_post_send - Posts a list of work requests to the send queue of
3749 * @qp: The QP to post the work request on.
3750 * @send_wr: A list of work requests to post on the send queue.
3751 * @bad_send_wr: On an immediate failure, this parameter will reference
3752 * the work request that failed to be posted on the QP.
3754 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3755 * error is returned, the QP state shall not be affected,
3756 * ib_post_send() will return an immediate error after queueing any
3757 * earlier work requests in the list.
3759 static inline int ib_post_send(struct ib_qp *qp,
3760 const struct ib_send_wr *send_wr,
3761 const struct ib_send_wr **bad_send_wr)
3763 const struct ib_send_wr *dummy;
3765 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3769 * ib_post_recv - Posts a list of work requests to the receive queue of
3771 * @qp: The QP to post the work request on.
3772 * @recv_wr: A list of work requests to post on the receive queue.
3773 * @bad_recv_wr: On an immediate failure, this parameter will reference
3774 * the work request that failed to be posted on the QP.
3776 static inline int ib_post_recv(struct ib_qp *qp,
3777 const struct ib_recv_wr *recv_wr,
3778 const struct ib_recv_wr **bad_recv_wr)
3780 const struct ib_recv_wr *dummy;
3782 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3785 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3786 int comp_vector, enum ib_poll_context poll_ctx,
3787 const char *caller);
3788 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3789 int nr_cqe, int comp_vector,
3790 enum ib_poll_context poll_ctx)
3792 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3796 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3797 int nr_cqe, enum ib_poll_context poll_ctx,
3798 const char *caller);
3801 * ib_alloc_cq_any: Allocate kernel CQ
3802 * @dev: The IB device
3803 * @private: Private data attached to the CQE
3804 * @nr_cqe: Number of CQEs in the CQ
3805 * @poll_ctx: Context used for polling the CQ
3807 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3808 void *private, int nr_cqe,
3809 enum ib_poll_context poll_ctx)
3811 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3815 void ib_free_cq(struct ib_cq *cq);
3816 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3819 * ib_create_cq - Creates a CQ on the specified device.
3820 * @device: The device on which to create the CQ.
3821 * @comp_handler: A user-specified callback that is invoked when a
3822 * completion event occurs on the CQ.
3823 * @event_handler: A user-specified callback that is invoked when an
3824 * asynchronous event not associated with a completion occurs on the CQ.
3825 * @cq_context: Context associated with the CQ returned to the user via
3826 * the associated completion and event handlers.
3827 * @cq_attr: The attributes the CQ should be created upon.
3829 * Users can examine the cq structure to determine the actual CQ size.
3831 struct ib_cq *__ib_create_cq(struct ib_device *device,
3832 ib_comp_handler comp_handler,
3833 void (*event_handler)(struct ib_event *, void *),
3835 const struct ib_cq_init_attr *cq_attr,
3836 const char *caller);
3837 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3838 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3841 * ib_resize_cq - Modifies the capacity of the CQ.
3842 * @cq: The CQ to resize.
3843 * @cqe: The minimum size of the CQ.
3845 * Users can examine the cq structure to determine the actual CQ size.
3847 int ib_resize_cq(struct ib_cq *cq, int cqe);
3850 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3851 * @cq: The CQ to modify.
3852 * @cq_count: number of CQEs that will trigger an event
3853 * @cq_period: max period of time in usec before triggering an event
3856 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3859 * ib_destroy_cq_user - Destroys the specified CQ.
3860 * @cq: The CQ to destroy.
3861 * @udata: Valid user data or NULL for kernel objects
3863 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3866 * ib_destroy_cq - Destroys the specified kernel CQ.
3867 * @cq: The CQ to destroy.
3869 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3871 static inline void ib_destroy_cq(struct ib_cq *cq)
3873 int ret = ib_destroy_cq_user(cq, NULL);
3875 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3879 * ib_poll_cq - poll a CQ for completion(s)
3880 * @cq:the CQ being polled
3881 * @num_entries:maximum number of completions to return
3882 * @wc:array of at least @num_entries &struct ib_wc where completions
3885 * Poll a CQ for (possibly multiple) completions. If the return value
3886 * is < 0, an error occurred. If the return value is >= 0, it is the
3887 * number of completions returned. If the return value is
3888 * non-negative and < num_entries, then the CQ was emptied.
3890 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3893 return cq->device->ops.poll_cq(cq, num_entries, wc);
3897 * ib_req_notify_cq - Request completion notification on a CQ.
3898 * @cq: The CQ to generate an event for.
3900 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3901 * to request an event on the next solicited event or next work
3902 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3903 * may also be |ed in to request a hint about missed events, as
3907 * < 0 means an error occurred while requesting notification
3908 * == 0 means notification was requested successfully, and if
3909 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3910 * were missed and it is safe to wait for another event. In
3911 * this case is it guaranteed that any work completions added
3912 * to the CQ since the last CQ poll will trigger a completion
3913 * notification event.
3914 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3915 * in. It means that the consumer must poll the CQ again to
3916 * make sure it is empty to avoid missing an event because of a
3917 * race between requesting notification and an entry being
3918 * added to the CQ. This return value means it is possible
3919 * (but not guaranteed) that a work completion has been added
3920 * to the CQ since the last poll without triggering a
3921 * completion notification event.
3923 static inline int ib_req_notify_cq(struct ib_cq *cq,
3924 enum ib_cq_notify_flags flags)
3926 return cq->device->ops.req_notify_cq(cq, flags);
3929 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3930 int comp_vector_hint,
3931 enum ib_poll_context poll_ctx);
3933 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3936 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3937 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3938 * address into the dma address.
3940 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3942 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3946 * ib_dma_mapping_error - check a DMA addr for error
3947 * @dev: The device for which the dma_addr was created
3948 * @dma_addr: The DMA address to check
3950 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3952 if (ib_uses_virt_dma(dev))
3954 return dma_mapping_error(dev->dma_device, dma_addr);
3958 * ib_dma_map_single - Map a kernel virtual address to DMA address
3959 * @dev: The device for which the dma_addr is to be created
3960 * @cpu_addr: The kernel virtual address
3961 * @size: The size of the region in bytes
3962 * @direction: The direction of the DMA
3964 static inline u64 ib_dma_map_single(struct ib_device *dev,
3965 void *cpu_addr, size_t size,
3966 enum dma_data_direction direction)
3968 if (ib_uses_virt_dma(dev))
3969 return (uintptr_t)cpu_addr;
3970 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3974 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3975 * @dev: The device for which the DMA address was created
3976 * @addr: The DMA address
3977 * @size: The size of the region in bytes
3978 * @direction: The direction of the DMA
3980 static inline void ib_dma_unmap_single(struct ib_device *dev,
3981 u64 addr, size_t size,
3982 enum dma_data_direction direction)
3984 if (!ib_uses_virt_dma(dev))
3985 dma_unmap_single(dev->dma_device, addr, size, direction);
3989 * ib_dma_map_page - Map a physical page to DMA address
3990 * @dev: The device for which the dma_addr is to be created
3991 * @page: The page to be mapped
3992 * @offset: The offset within the page
3993 * @size: The size of the region in bytes
3994 * @direction: The direction of the DMA
3996 static inline u64 ib_dma_map_page(struct ib_device *dev,
3998 unsigned long offset,
4000 enum dma_data_direction direction)
4002 if (ib_uses_virt_dma(dev))
4003 return (uintptr_t)(page_address(page) + offset);
4004 return dma_map_page(dev->dma_device, page, offset, size, direction);
4008 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4009 * @dev: The device for which the DMA address was created
4010 * @addr: The DMA address
4011 * @size: The size of the region in bytes
4012 * @direction: The direction of the DMA
4014 static inline void ib_dma_unmap_page(struct ib_device *dev,
4015 u64 addr, size_t size,
4016 enum dma_data_direction direction)
4018 if (!ib_uses_virt_dma(dev))
4019 dma_unmap_page(dev->dma_device, addr, size, direction);
4022 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4023 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4024 struct scatterlist *sg, int nents,
4025 enum dma_data_direction direction,
4026 unsigned long dma_attrs)
4028 if (ib_uses_virt_dma(dev))
4029 return ib_dma_virt_map_sg(dev, sg, nents);
4030 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4034 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4035 struct scatterlist *sg, int nents,
4036 enum dma_data_direction direction,
4037 unsigned long dma_attrs)
4039 if (!ib_uses_virt_dma(dev))
4040 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4045 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4046 * @dev: The device for which the DMA addresses are to be created
4047 * @sg: The array of scatter/gather entries
4048 * @nents: The number of scatter/gather entries
4049 * @direction: The direction of the DMA
4051 static inline int ib_dma_map_sg(struct ib_device *dev,
4052 struct scatterlist *sg, int nents,
4053 enum dma_data_direction direction)
4055 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4059 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4060 * @dev: The device for which the DMA addresses were created
4061 * @sg: The array of scatter/gather entries
4062 * @nents: The number of scatter/gather entries
4063 * @direction: The direction of the DMA
4065 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4066 struct scatterlist *sg, int nents,
4067 enum dma_data_direction direction)
4069 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4073 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4074 * @dev: The device to query
4076 * The returned value represents a size in bytes.
4078 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4080 if (ib_uses_virt_dma(dev))
4082 return dma_get_max_seg_size(dev->dma_device);
4086 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4087 * @dev: The device for which the DMA address was created
4088 * @addr: The DMA address
4089 * @size: The size of the region in bytes
4090 * @dir: The direction of the DMA
4092 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4095 enum dma_data_direction dir)
4097 if (!ib_uses_virt_dma(dev))
4098 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4102 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4103 * @dev: The device for which the DMA address was created
4104 * @addr: The DMA address
4105 * @size: The size of the region in bytes
4106 * @dir: The direction of the DMA
4108 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4111 enum dma_data_direction dir)
4113 if (!ib_uses_virt_dma(dev))
4114 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4117 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4118 * space. This function should be called when 'current' is the owning MM.
4120 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4121 u64 virt_addr, int mr_access_flags);
4123 /* ib_advise_mr - give an advice about an address range in a memory region */
4124 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4125 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4127 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4128 * HCA translation table.
4129 * @mr: The memory region to deregister.
4130 * @udata: Valid user data or NULL for kernel object
4132 * This function can fail, if the memory region has memory windows bound to it.
4134 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4137 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4138 * HCA translation table.
4139 * @mr: The memory region to deregister.
4141 * This function can fail, if the memory region has memory windows bound to it.
4143 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4145 static inline int ib_dereg_mr(struct ib_mr *mr)
4147 return ib_dereg_mr_user(mr, NULL);
4150 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4153 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4154 u32 max_num_data_sg,
4155 u32 max_num_meta_sg);
4158 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4160 * @mr - struct ib_mr pointer to be updated.
4161 * @newkey - new key to be used.
4163 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4165 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4166 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4170 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4171 * for calculating a new rkey for type 2 memory windows.
4172 * @rkey - the rkey to increment.
4174 static inline u32 ib_inc_rkey(u32 rkey)
4176 const u32 mask = 0x000000ff;
4177 return ((rkey + 1) & mask) | (rkey & ~mask);
4181 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4182 * @qp: QP to attach to the multicast group. The QP must be type
4184 * @gid: Multicast group GID.
4185 * @lid: Multicast group LID in host byte order.
4187 * In order to send and receive multicast packets, subnet
4188 * administration must have created the multicast group and configured
4189 * the fabric appropriately. The port associated with the specified
4190 * QP must also be a member of the multicast group.
4192 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4195 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4196 * @qp: QP to detach from the multicast group.
4197 * @gid: Multicast group GID.
4198 * @lid: Multicast group LID in host byte order.
4200 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4202 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4203 struct inode *inode, struct ib_udata *udata);
4204 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4206 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4210 * Local write permission is required if remote write or
4211 * remote atomic permission is also requested.
4213 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4214 !(flags & IB_ACCESS_LOCAL_WRITE))
4217 if (flags & ~IB_ACCESS_SUPPORTED)
4220 if (flags & IB_ACCESS_ON_DEMAND &&
4221 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4226 static inline bool ib_access_writable(int access_flags)
4229 * We have writable memory backing the MR if any of the following
4230 * access flags are set. "Local write" and "remote write" obviously
4231 * require write access. "Remote atomic" can do things like fetch and
4232 * add, which will modify memory, and "MW bind" can change permissions
4233 * by binding a window.
4235 return access_flags &
4236 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4237 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4241 * ib_check_mr_status: lightweight check of MR status.
4242 * This routine may provide status checks on a selected
4243 * ib_mr. first use is for signature status check.
4245 * @mr: A memory region.
4246 * @check_mask: Bitmask of which checks to perform from
4247 * ib_mr_status_check enumeration.
4248 * @mr_status: The container of relevant status checks.
4249 * failed checks will be indicated in the status bitmask
4250 * and the relevant info shall be in the error item.
4252 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4253 struct ib_mr_status *mr_status);
4256 * ib_device_try_get: Hold a registration lock
4257 * device: The device to lock
4259 * A device under an active registration lock cannot become unregistered. It
4260 * is only possible to obtain a registration lock on a device that is fully
4261 * registered, otherwise this function returns false.
4263 * The registration lock is only necessary for actions which require the
4264 * device to still be registered. Uses that only require the device pointer to
4265 * be valid should use get_device(&ibdev->dev) to hold the memory.
4268 static inline bool ib_device_try_get(struct ib_device *dev)
4270 return refcount_inc_not_zero(&dev->refcount);
4273 void ib_device_put(struct ib_device *device);
4274 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4275 enum rdma_driver_id driver_id);
4276 struct ib_device *ib_device_get_by_name(const char *name,
4277 enum rdma_driver_id driver_id);
4278 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4279 u16 pkey, const union ib_gid *gid,
4280 const struct sockaddr *addr);
4281 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4283 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4285 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4286 struct ib_wq_init_attr *init_attr);
4287 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4289 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4290 unsigned int *sg_offset, unsigned int page_size);
4291 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4292 int data_sg_nents, unsigned int *data_sg_offset,
4293 struct scatterlist *meta_sg, int meta_sg_nents,
4294 unsigned int *meta_sg_offset, unsigned int page_size);
4297 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4298 unsigned int *sg_offset, unsigned int page_size)
4302 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4308 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4309 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4311 void ib_drain_rq(struct ib_qp *qp);
4312 void ib_drain_sq(struct ib_qp *qp);
4313 void ib_drain_qp(struct ib_qp *qp);
4315 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4318 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4320 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4321 return attr->roce.dmac;
4325 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4327 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4328 attr->ib.dlid = (u16)dlid;
4329 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4330 attr->opa.dlid = dlid;
4333 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4335 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4336 return attr->ib.dlid;
4337 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4338 return attr->opa.dlid;
4342 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4347 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4352 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4355 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4356 attr->ib.src_path_bits = src_path_bits;
4357 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4358 attr->opa.src_path_bits = src_path_bits;
4361 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4363 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4364 return attr->ib.src_path_bits;
4365 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4366 return attr->opa.src_path_bits;
4370 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4373 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4374 attr->opa.make_grd = make_grd;
4377 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4379 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4380 return attr->opa.make_grd;
4384 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4386 attr->port_num = port_num;
4389 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4391 return attr->port_num;
4394 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4397 attr->static_rate = static_rate;
4400 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4402 return attr->static_rate;
4405 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4406 enum ib_ah_flags flag)
4408 attr->ah_flags = flag;
4411 static inline enum ib_ah_flags
4412 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4414 return attr->ah_flags;
4417 static inline const struct ib_global_route
4418 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4423 /*To retrieve and modify the grh */
4424 static inline struct ib_global_route
4425 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4430 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4432 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4434 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4437 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4440 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4442 grh->dgid.global.subnet_prefix = prefix;
4445 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4448 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4450 grh->dgid.global.interface_id = if_id;
4453 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4454 union ib_gid *dgid, u32 flow_label,
4455 u8 sgid_index, u8 hop_limit,
4458 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4460 attr->ah_flags = IB_AH_GRH;
4463 grh->flow_label = flow_label;
4464 grh->sgid_index = sgid_index;
4465 grh->hop_limit = hop_limit;
4466 grh->traffic_class = traffic_class;
4467 grh->sgid_attr = NULL;
4470 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4471 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4472 u32 flow_label, u8 hop_limit, u8 traffic_class,
4473 const struct ib_gid_attr *sgid_attr);
4474 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4475 const struct rdma_ah_attr *src);
4476 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4477 const struct rdma_ah_attr *new);
4478 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4481 * rdma_ah_find_type - Return address handle type.
4483 * @dev: Device to be checked
4484 * @port_num: Port number
4486 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4489 if (rdma_protocol_roce(dev, port_num))
4490 return RDMA_AH_ATTR_TYPE_ROCE;
4491 if (rdma_protocol_ib(dev, port_num)) {
4492 if (rdma_cap_opa_ah(dev, port_num))
4493 return RDMA_AH_ATTR_TYPE_OPA;
4494 return RDMA_AH_ATTR_TYPE_IB;
4497 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4501 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4502 * In the current implementation the only way to get
4503 * get the 32bit lid is from other sources for OPA.
4504 * For IB, lids will always be 16bits so cast the
4505 * value accordingly.
4509 static inline u16 ib_lid_cpu16(u32 lid)
4511 WARN_ON_ONCE(lid & 0xFFFF0000);
4516 * ib_lid_be16 - Return lid in 16bit BE encoding.
4520 static inline __be16 ib_lid_be16(u32 lid)
4522 WARN_ON_ONCE(lid & 0xFFFF0000);
4523 return cpu_to_be16((u16)lid);
4527 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4529 * @device: the rdma device
4530 * @comp_vector: index of completion vector
4532 * Returns NULL on failure, otherwise a corresponding cpu map of the
4533 * completion vector (returns all-cpus map if the device driver doesn't
4534 * implement get_vector_affinity).
4536 static inline const struct cpumask *
4537 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4539 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4540 !device->ops.get_vector_affinity)
4543 return device->ops.get_vector_affinity(device, comp_vector);
4548 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4549 * and add their gids, as needed, to the relevant RoCE devices.
4551 * @device: the rdma device
4553 void rdma_roce_rescan_device(struct ib_device *ibdev);
4555 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4557 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4559 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4560 enum rdma_netdev_t type, const char *name,
4561 unsigned char name_assign_type,
4562 void (*setup)(struct net_device *));
4564 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4565 enum rdma_netdev_t type, const char *name,
4566 unsigned char name_assign_type,
4567 void (*setup)(struct net_device *),
4568 struct net_device *netdev);
4571 * rdma_set_device_sysfs_group - Set device attributes group to have
4572 * driver specific sysfs entries at
4573 * for infiniband class.
4575 * @device: device pointer for which attributes to be created
4576 * @group: Pointer to group which should be added when device
4577 * is registered with sysfs.
4578 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4579 * group per device to have sysfs attributes.
4581 * NOTE: New drivers should not make use of this API; instead new device
4582 * parameter should be exposed via netlink command. This API and mechanism
4583 * exist only for existing drivers.
4586 rdma_set_device_sysfs_group(struct ib_device *dev,
4587 const struct attribute_group *group)
4589 dev->groups[1] = group;
4593 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4595 * @device: device pointer for which ib_device pointer to retrieve
4597 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4600 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4602 struct ib_core_device *coredev =
4603 container_of(device, struct ib_core_device, dev);
4605 return coredev->owner;
4609 * ibdev_to_node - return the NUMA node for a given ib_device
4610 * @dev: device to get the NUMA node for.
4612 static inline int ibdev_to_node(struct ib_device *ibdev)
4614 struct device *parent = ibdev->dev.parent;
4617 return NUMA_NO_NODE;
4618 return dev_to_node(parent);
4622 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4623 * ib_device holder structure from device pointer.
4625 * NOTE: New drivers should not make use of this API; This API is only for
4626 * existing drivers who have exposed sysfs entries using
4627 * rdma_set_device_sysfs_group().
4629 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4630 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4632 bool rdma_dev_access_netns(const struct ib_device *device,
4633 const struct net *net);
4635 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4636 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4637 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4640 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4643 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4644 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4647 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4649 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4651 fl_low ^= fl_high >> 14;
4652 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4656 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4657 * local and remote qpn values
4659 * This function folded the multiplication results of two qpns, 24 bit each,
4660 * fields, and converts it to a 20 bit results.
4662 * This function will create symmetric flow_label value based on the local
4663 * and remote qpn values. this will allow both the requester and responder
4664 * to calculate the same flow_label for a given connection.
4666 * This helper function should be used by driver in case the upper layer
4667 * provide a zero flow_label value. This is to improve entropy of RDMA
4668 * traffic in the network.
4670 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4672 u64 v = (u64)lqpn * rqpn;
4677 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4680 const struct ib_port_immutable*
4681 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4682 #endif /* IB_VERBS_H */