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 extern struct workqueue_struct *ib_wq;
55 extern struct workqueue_struct *ib_comp_wq;
56 extern struct workqueue_struct *ib_comp_unbound_wq;
61 void ibdev_printk(const char *level, const struct ib_device *ibdev,
62 const char *format, ...);
64 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
66 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
68 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
70 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
72 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
74 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
76 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
78 #if defined(CONFIG_DYNAMIC_DEBUG) || \
79 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
80 #define ibdev_dbg(__dev, format, args...) \
81 dynamic_ibdev_dbg(__dev, format, ##args)
85 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
90 static DEFINE_RATELIMIT_STATE(_rs, \
91 DEFAULT_RATELIMIT_INTERVAL, \
92 DEFAULT_RATELIMIT_BURST); \
93 if (__ratelimit(&_rs)) \
94 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
97 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
98 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
99 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
112 #if defined(CONFIG_DYNAMIC_DEBUG) || \
113 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
114 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
115 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
117 static DEFINE_RATELIMIT_STATE(_rs, \
118 DEFAULT_RATELIMIT_INTERVAL, \
119 DEFAULT_RATELIMIT_BURST); \
120 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
121 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
122 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
126 __printf(2, 3) __cold
128 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
134 __be64 subnet_prefix;
139 extern union ib_gid zgid;
142 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
143 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
144 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
148 #define ROCE_V2_UDP_DPORT 4791
150 struct net_device __rcu *ndev;
151 struct ib_device *device;
153 enum ib_gid_type gid_type;
159 /* set the local administered indication */
160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
163 enum rdma_transport_type {
165 RDMA_TRANSPORT_IWARP,
166 RDMA_TRANSPORT_USNIC,
167 RDMA_TRANSPORT_USNIC_UDP,
168 RDMA_TRANSPORT_UNSPECIFIED,
171 enum rdma_protocol_type {
175 RDMA_PROTOCOL_USNIC_UDP
178 __attribute_const__ enum rdma_transport_type
179 rdma_node_get_transport(unsigned int node_type);
181 enum rdma_network_type {
183 RDMA_NETWORK_ROCE_V1,
188 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
190 if (network_type == RDMA_NETWORK_IPV4 ||
191 network_type == RDMA_NETWORK_IPV6)
192 return IB_GID_TYPE_ROCE_UDP_ENCAP;
193 else if (network_type == RDMA_NETWORK_ROCE_V1)
194 return IB_GID_TYPE_ROCE;
196 return IB_GID_TYPE_IB;
199 static inline enum rdma_network_type
200 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
202 if (attr->gid_type == IB_GID_TYPE_IB)
203 return RDMA_NETWORK_IB;
205 if (attr->gid_type == IB_GID_TYPE_ROCE)
206 return RDMA_NETWORK_ROCE_V1;
208 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
209 return RDMA_NETWORK_IPV4;
211 return RDMA_NETWORK_IPV6;
214 enum rdma_link_layer {
215 IB_LINK_LAYER_UNSPECIFIED,
216 IB_LINK_LAYER_INFINIBAND,
217 IB_LINK_LAYER_ETHERNET,
220 enum ib_device_cap_flags {
221 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
222 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
223 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
224 IB_DEVICE_RAW_MULTI = (1 << 3),
225 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
226 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
227 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
228 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
229 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
230 /* Not in use, former INIT_TYPE = (1 << 9),*/
231 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
232 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
233 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
234 IB_DEVICE_SRQ_RESIZE = (1 << 13),
235 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
238 * This device supports a per-device lkey or stag that can be
239 * used without performing a memory registration for the local
240 * memory. Note that ULPs should never check this flag, but
241 * instead of use the local_dma_lkey flag in the ib_pd structure,
242 * which will always contain a usable lkey.
244 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
245 /* Reserved, old SEND_W_INV = (1 << 16),*/
246 IB_DEVICE_MEM_WINDOW = (1 << 17),
248 * Devices should set IB_DEVICE_UD_IP_SUM if they support
249 * insertion of UDP and TCP checksum on outgoing UD IPoIB
250 * messages and can verify the validity of checksum for
251 * incoming messages. Setting this flag implies that the
252 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
254 IB_DEVICE_UD_IP_CSUM = (1 << 18),
255 IB_DEVICE_UD_TSO = (1 << 19),
256 IB_DEVICE_XRC = (1 << 20),
259 * This device supports the IB "base memory management extension",
260 * which includes support for fast registrations (IB_WR_REG_MR,
261 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
262 * also be set by any iWarp device which must support FRs to comply
263 * to the iWarp verbs spec. iWarp devices also support the
264 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
267 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
268 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
269 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
270 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
271 IB_DEVICE_RC_IP_CSUM = (1 << 25),
272 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
273 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
275 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
276 * support execution of WQEs that involve synchronization
277 * of I/O operations with single completion queue managed
280 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
281 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
282 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
283 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
284 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
285 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
286 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
287 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
288 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
289 /* The device supports padding incoming writes to cacheline. */
290 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
291 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
300 enum ib_odp_general_cap_bits {
301 IB_ODP_SUPPORT = 1 << 0,
302 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
305 enum ib_odp_transport_cap_bits {
306 IB_ODP_SUPPORT_SEND = 1 << 0,
307 IB_ODP_SUPPORT_RECV = 1 << 1,
308 IB_ODP_SUPPORT_WRITE = 1 << 2,
309 IB_ODP_SUPPORT_READ = 1 << 3,
310 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
311 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
315 uint64_t general_caps;
317 uint32_t rc_odp_caps;
318 uint32_t uc_odp_caps;
319 uint32_t ud_odp_caps;
320 uint32_t xrc_odp_caps;
321 } per_transport_caps;
325 /* Corresponding bit will be set if qp type from
326 * 'enum ib_qp_type' is supported, e.g.
327 * supported_qpts |= 1 << IB_QPT_UD
330 u32 max_rwq_indirection_tables;
331 u32 max_rwq_indirection_table_size;
334 enum ib_tm_cap_flags {
335 /* Support tag matching with rendezvous offload for RC transport */
336 IB_TM_CAP_RNDV_RC = 1 << 0,
340 /* Max size of RNDV header */
341 u32 max_rndv_hdr_size;
342 /* Max number of entries in tag matching list */
344 /* From enum ib_tm_cap_flags */
346 /* Max number of outstanding list operations */
348 /* Max number of SGE in tag matching entry */
352 struct ib_cq_init_attr {
358 enum ib_cq_attr_mask {
359 IB_CQ_MODERATE = 1 << 0,
363 u16 max_cq_moderation_count;
364 u16 max_cq_moderation_period;
367 struct ib_dm_mr_attr {
373 struct ib_dm_alloc_attr {
379 struct ib_device_attr {
381 __be64 sys_image_guid;
389 u64 device_cap_flags;
400 int max_qp_init_rd_atom;
401 int max_ee_init_rd_atom;
402 enum ib_atomic_cap atomic_cap;
403 enum ib_atomic_cap masked_atomic_cap;
410 int max_mcast_qp_attach;
411 int max_total_mcast_qp_attach;
416 unsigned int max_fast_reg_page_list_len;
417 unsigned int max_pi_fast_reg_page_list_len;
419 u8 local_ca_ack_delay;
422 struct ib_odp_caps odp_caps;
423 uint64_t timestamp_mask;
424 uint64_t hca_core_clock; /* in KHZ */
425 struct ib_rss_caps rss_caps;
427 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
428 struct ib_tm_caps tm_caps;
429 struct ib_cq_caps cq_caps;
431 /* Max entries for sgl for optimized performance per READ */
448 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
451 case IB_MTU_256: return 256;
452 case IB_MTU_512: return 512;
453 case IB_MTU_1024: return 1024;
454 case IB_MTU_2048: return 2048;
455 case IB_MTU_4096: return 4096;
460 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
464 else if (mtu >= 2048)
466 else if (mtu >= 1024)
474 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
482 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
486 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
489 return OPA_MTU_10240;
490 else if (mtu >= 8192)
493 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
502 IB_PORT_ACTIVE_DEFER = 5
505 enum ib_port_phys_state {
506 IB_PORT_PHYS_STATE_SLEEP = 1,
507 IB_PORT_PHYS_STATE_POLLING = 2,
508 IB_PORT_PHYS_STATE_DISABLED = 3,
509 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
510 IB_PORT_PHYS_STATE_LINK_UP = 5,
511 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
512 IB_PORT_PHYS_STATE_PHY_TEST = 7,
523 static inline int ib_width_enum_to_int(enum ib_port_width width)
526 case IB_WIDTH_1X: return 1;
527 case IB_WIDTH_2X: return 2;
528 case IB_WIDTH_4X: return 4;
529 case IB_WIDTH_8X: return 8;
530 case IB_WIDTH_12X: return 12;
547 * struct rdma_hw_stats
548 * @lock - Mutex to protect parallel write access to lifespan and values
549 * of counters, which are 64bits and not guaranteeed to be written
550 * atomicaly on 32bits systems.
551 * @timestamp - Used by the core code to track when the last update was
552 * @lifespan - Used by the core code to determine how old the counters
553 * should be before being updated again. Stored in jiffies, defaults
554 * to 10 milliseconds, drivers can override the default be specifying
555 * their own value during their allocation routine.
556 * @name - Array of pointers to static names used for the counters in
558 * @num_counters - How many hardware counters there are. If name is
559 * shorter than this number, a kernel oops will result. Driver authors
560 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
561 * in their code to prevent this.
562 * @value - Array of u64 counters that are accessed by the sysfs code and
563 * filled in by the drivers get_stats routine
565 struct rdma_hw_stats {
566 struct mutex lock; /* Protect lifespan and values[] */
567 unsigned long timestamp;
568 unsigned long lifespan;
569 const char * const *names;
574 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
576 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
578 * @names - Array of static const char *
579 * @num_counters - How many elements in array
580 * @lifespan - How many milliseconds between updates
582 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
583 const char * const *names, int num_counters,
584 unsigned long lifespan)
586 struct rdma_hw_stats *stats;
588 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
592 stats->names = names;
593 stats->num_counters = num_counters;
594 stats->lifespan = msecs_to_jiffies(lifespan);
600 /* Define bits for the various functionality this port needs to be supported by
603 /* Management 0x00000FFF */
604 #define RDMA_CORE_CAP_IB_MAD 0x00000001
605 #define RDMA_CORE_CAP_IB_SMI 0x00000002
606 #define RDMA_CORE_CAP_IB_CM 0x00000004
607 #define RDMA_CORE_CAP_IW_CM 0x00000008
608 #define RDMA_CORE_CAP_IB_SA 0x00000010
609 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
611 /* Address format 0x000FF000 */
612 #define RDMA_CORE_CAP_AF_IB 0x00001000
613 #define RDMA_CORE_CAP_ETH_AH 0x00002000
614 #define RDMA_CORE_CAP_OPA_AH 0x00004000
615 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
617 /* Protocol 0xFFF00000 */
618 #define RDMA_CORE_CAP_PROT_IB 0x00100000
619 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
620 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
621 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
622 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
623 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
625 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
626 | RDMA_CORE_CAP_PROT_ROCE \
627 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
629 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
630 | RDMA_CORE_CAP_IB_MAD \
631 | RDMA_CORE_CAP_IB_SMI \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_IB_SA \
634 | RDMA_CORE_CAP_AF_IB)
635 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
636 | RDMA_CORE_CAP_IB_MAD \
637 | RDMA_CORE_CAP_IB_CM \
638 | RDMA_CORE_CAP_AF_IB \
639 | RDMA_CORE_CAP_ETH_AH)
640 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
641 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
642 | RDMA_CORE_CAP_IB_MAD \
643 | RDMA_CORE_CAP_IB_CM \
644 | RDMA_CORE_CAP_AF_IB \
645 | RDMA_CORE_CAP_ETH_AH)
646 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
647 | RDMA_CORE_CAP_IW_CM)
648 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
649 | RDMA_CORE_CAP_OPA_MAD)
651 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
653 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
655 struct ib_port_attr {
657 enum ib_port_state state;
659 enum ib_mtu active_mtu;
662 unsigned int ip_gids:1;
663 /* This is the value from PortInfo CapabilityMask, defined by IBA */
682 enum ib_device_modify_flags {
683 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
684 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
687 #define IB_DEVICE_NODE_DESC_MAX 64
689 struct ib_device_modify {
691 char node_desc[IB_DEVICE_NODE_DESC_MAX];
694 enum ib_port_modify_flags {
695 IB_PORT_SHUTDOWN = 1,
696 IB_PORT_INIT_TYPE = (1<<2),
697 IB_PORT_RESET_QKEY_CNTR = (1<<3),
698 IB_PORT_OPA_MASK_CHG = (1<<4)
701 struct ib_port_modify {
702 u32 set_port_cap_mask;
703 u32 clr_port_cap_mask;
711 IB_EVENT_QP_ACCESS_ERR,
715 IB_EVENT_PATH_MIG_ERR,
716 IB_EVENT_DEVICE_FATAL,
717 IB_EVENT_PORT_ACTIVE,
720 IB_EVENT_PKEY_CHANGE,
723 IB_EVENT_SRQ_LIMIT_REACHED,
724 IB_EVENT_QP_LAST_WQE_REACHED,
725 IB_EVENT_CLIENT_REREGISTER,
730 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
733 struct ib_device *device;
741 enum ib_event_type event;
744 struct ib_event_handler {
745 struct ib_device *device;
746 void (*handler)(struct ib_event_handler *, struct ib_event *);
747 struct list_head list;
750 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
752 (_ptr)->device = _device; \
753 (_ptr)->handler = _handler; \
754 INIT_LIST_HEAD(&(_ptr)->list); \
757 struct ib_global_route {
758 const struct ib_gid_attr *sgid_attr;
767 __be32 version_tclass_flow;
775 union rdma_network_hdr {
778 /* The IB spec states that if it's IPv4, the header
779 * is located in the last 20 bytes of the header.
782 struct iphdr roce4grh;
786 #define IB_QPN_MASK 0xFFFFFF
789 IB_MULTICAST_QPN = 0xffffff
792 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
793 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
800 IB_RATE_PORT_CURRENT = 0,
801 IB_RATE_2_5_GBPS = 2,
809 IB_RATE_120_GBPS = 10,
810 IB_RATE_14_GBPS = 11,
811 IB_RATE_56_GBPS = 12,
812 IB_RATE_112_GBPS = 13,
813 IB_RATE_168_GBPS = 14,
814 IB_RATE_25_GBPS = 15,
815 IB_RATE_100_GBPS = 16,
816 IB_RATE_200_GBPS = 17,
817 IB_RATE_300_GBPS = 18,
818 IB_RATE_28_GBPS = 19,
819 IB_RATE_50_GBPS = 20,
820 IB_RATE_400_GBPS = 21,
821 IB_RATE_600_GBPS = 22,
825 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
826 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
827 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
828 * @rate: rate to convert.
830 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
833 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
834 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
835 * @rate: rate to convert.
837 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
841 * enum ib_mr_type - memory region type
842 * @IB_MR_TYPE_MEM_REG: memory region that is used for
843 * normal registration
844 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
845 * register any arbitrary sg lists (without
846 * the normal mr constraints - see
848 * @IB_MR_TYPE_DM: memory region that is used for device
849 * memory registration
850 * @IB_MR_TYPE_USER: memory region that is used for the user-space
852 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
853 * without address translations (VA=PA)
854 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
855 * data integrity operations
863 IB_MR_TYPE_INTEGRITY,
866 enum ib_mr_status_check {
867 IB_MR_CHECK_SIG_STATUS = 1,
871 * struct ib_mr_status - Memory region status container
873 * @fail_status: Bitmask of MR checks status. For each
874 * failed check a corresponding status bit is set.
875 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
878 struct ib_mr_status {
880 struct ib_sig_err sig_err;
884 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
886 * @mult: multiple to convert.
888 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
890 struct rdma_ah_init_attr {
891 struct rdma_ah_attr *ah_attr;
893 struct net_device *xmit_slave;
896 enum rdma_ah_attr_type {
897 RDMA_AH_ATTR_TYPE_UNDEFINED,
898 RDMA_AH_ATTR_TYPE_IB,
899 RDMA_AH_ATTR_TYPE_ROCE,
900 RDMA_AH_ATTR_TYPE_OPA,
908 struct roce_ah_attr {
918 struct rdma_ah_attr {
919 struct ib_global_route grh;
924 enum rdma_ah_attr_type type;
926 struct ib_ah_attr ib;
927 struct roce_ah_attr roce;
928 struct opa_ah_attr opa;
936 IB_WC_LOC_EEC_OP_ERR,
941 IB_WC_LOC_ACCESS_ERR,
942 IB_WC_REM_INV_REQ_ERR,
943 IB_WC_REM_ACCESS_ERR,
946 IB_WC_RNR_RETRY_EXC_ERR,
947 IB_WC_LOC_RDD_VIOL_ERR,
948 IB_WC_REM_INV_RD_REQ_ERR,
951 IB_WC_INV_EEC_STATE_ERR,
953 IB_WC_RESP_TIMEOUT_ERR,
957 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
960 IB_WC_SEND = IB_UVERBS_WC_SEND,
961 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
962 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
963 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
964 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
965 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
966 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
967 IB_WC_LSO = IB_UVERBS_WC_TSO,
969 IB_WC_MASKED_COMP_SWAP,
970 IB_WC_MASKED_FETCH_ADD,
972 * Set value of IB_WC_RECV so consumers can test if a completion is a
973 * receive by testing (opcode & IB_WC_RECV).
976 IB_WC_RECV_RDMA_WITH_IMM
981 IB_WC_WITH_IMM = (1<<1),
982 IB_WC_WITH_INVALIDATE = (1<<2),
983 IB_WC_IP_CSUM_OK = (1<<3),
984 IB_WC_WITH_SMAC = (1<<4),
985 IB_WC_WITH_VLAN = (1<<5),
986 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
992 struct ib_cqe *wr_cqe;
994 enum ib_wc_status status;
995 enum ib_wc_opcode opcode;
1001 u32 invalidate_rkey;
1009 u32 port_num; /* valid only for DR SMPs on switches */
1012 u8 network_hdr_type;
1015 enum ib_cq_notify_flags {
1016 IB_CQ_SOLICITED = 1 << 0,
1017 IB_CQ_NEXT_COMP = 1 << 1,
1018 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1019 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1023 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1024 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1025 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1028 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1030 return srq_type == IB_SRQT_XRC ||
1031 srq_type == IB_SRQT_TM;
1034 enum ib_srq_attr_mask {
1035 IB_SRQ_MAX_WR = 1 << 0,
1036 IB_SRQ_LIMIT = 1 << 1,
1039 struct ib_srq_attr {
1045 struct ib_srq_init_attr {
1046 void (*event_handler)(struct ib_event *, void *);
1048 struct ib_srq_attr attr;
1049 enum ib_srq_type srq_type;
1055 struct ib_xrcd *xrcd;
1070 u32 max_inline_data;
1073 * Maximum number of rdma_rw_ctx structures in flight at a time.
1074 * ib_create_qp() will calculate the right amount of neededed WRs
1075 * and MRs based on this.
1087 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1088 * here (and in that order) since the MAD layer uses them as
1089 * indices into a 2-entry table.
1094 IB_QPT_RC = IB_UVERBS_QPT_RC,
1095 IB_QPT_UC = IB_UVERBS_QPT_UC,
1096 IB_QPT_UD = IB_UVERBS_QPT_UD,
1098 IB_QPT_RAW_ETHERTYPE,
1099 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1100 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1101 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1103 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1104 /* Reserve a range for qp types internal to the low level driver.
1105 * These qp types will not be visible at the IB core layer, so the
1106 * IB_QPT_MAX usages should not be affected in the core layer
1108 IB_QPT_RESERVED1 = 0x1000,
1120 enum ib_qp_create_flags {
1121 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1122 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1123 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1124 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1125 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1126 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1127 IB_QP_CREATE_NETIF_QP = 1 << 5,
1128 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1129 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1130 IB_QP_CREATE_SCATTER_FCS =
1131 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1132 IB_QP_CREATE_CVLAN_STRIPPING =
1133 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1134 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1135 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1136 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1137 /* reserve bits 26-31 for low level drivers' internal use */
1138 IB_QP_CREATE_RESERVED_START = 1 << 26,
1139 IB_QP_CREATE_RESERVED_END = 1 << 31,
1143 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1144 * callback to destroy the passed in QP.
1147 struct ib_qp_init_attr {
1148 /* Consumer's event_handler callback must not block */
1149 void (*event_handler)(struct ib_event *, void *);
1152 struct ib_cq *send_cq;
1153 struct ib_cq *recv_cq;
1155 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1156 struct ib_qp_cap cap;
1157 enum ib_sig_type sq_sig_type;
1158 enum ib_qp_type qp_type;
1162 * Only needed for special QP types, or when using the RW API.
1165 struct ib_rwq_ind_table *rwq_ind_tbl;
1169 struct ib_qp_open_attr {
1170 void (*event_handler)(struct ib_event *, void *);
1173 enum ib_qp_type qp_type;
1176 enum ib_rnr_timeout {
1177 IB_RNR_TIMER_655_36 = 0,
1178 IB_RNR_TIMER_000_01 = 1,
1179 IB_RNR_TIMER_000_02 = 2,
1180 IB_RNR_TIMER_000_03 = 3,
1181 IB_RNR_TIMER_000_04 = 4,
1182 IB_RNR_TIMER_000_06 = 5,
1183 IB_RNR_TIMER_000_08 = 6,
1184 IB_RNR_TIMER_000_12 = 7,
1185 IB_RNR_TIMER_000_16 = 8,
1186 IB_RNR_TIMER_000_24 = 9,
1187 IB_RNR_TIMER_000_32 = 10,
1188 IB_RNR_TIMER_000_48 = 11,
1189 IB_RNR_TIMER_000_64 = 12,
1190 IB_RNR_TIMER_000_96 = 13,
1191 IB_RNR_TIMER_001_28 = 14,
1192 IB_RNR_TIMER_001_92 = 15,
1193 IB_RNR_TIMER_002_56 = 16,
1194 IB_RNR_TIMER_003_84 = 17,
1195 IB_RNR_TIMER_005_12 = 18,
1196 IB_RNR_TIMER_007_68 = 19,
1197 IB_RNR_TIMER_010_24 = 20,
1198 IB_RNR_TIMER_015_36 = 21,
1199 IB_RNR_TIMER_020_48 = 22,
1200 IB_RNR_TIMER_030_72 = 23,
1201 IB_RNR_TIMER_040_96 = 24,
1202 IB_RNR_TIMER_061_44 = 25,
1203 IB_RNR_TIMER_081_92 = 26,
1204 IB_RNR_TIMER_122_88 = 27,
1205 IB_RNR_TIMER_163_84 = 28,
1206 IB_RNR_TIMER_245_76 = 29,
1207 IB_RNR_TIMER_327_68 = 30,
1208 IB_RNR_TIMER_491_52 = 31
1211 enum ib_qp_attr_mask {
1213 IB_QP_CUR_STATE = (1<<1),
1214 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1215 IB_QP_ACCESS_FLAGS = (1<<3),
1216 IB_QP_PKEY_INDEX = (1<<4),
1217 IB_QP_PORT = (1<<5),
1218 IB_QP_QKEY = (1<<6),
1220 IB_QP_PATH_MTU = (1<<8),
1221 IB_QP_TIMEOUT = (1<<9),
1222 IB_QP_RETRY_CNT = (1<<10),
1223 IB_QP_RNR_RETRY = (1<<11),
1224 IB_QP_RQ_PSN = (1<<12),
1225 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1226 IB_QP_ALT_PATH = (1<<14),
1227 IB_QP_MIN_RNR_TIMER = (1<<15),
1228 IB_QP_SQ_PSN = (1<<16),
1229 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1230 IB_QP_PATH_MIG_STATE = (1<<18),
1231 IB_QP_CAP = (1<<19),
1232 IB_QP_DEST_QPN = (1<<20),
1233 IB_QP_RESERVED1 = (1<<21),
1234 IB_QP_RESERVED2 = (1<<22),
1235 IB_QP_RESERVED3 = (1<<23),
1236 IB_QP_RESERVED4 = (1<<24),
1237 IB_QP_RATE_LIMIT = (1<<25),
1239 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1264 enum ib_qp_state qp_state;
1265 enum ib_qp_state cur_qp_state;
1266 enum ib_mtu path_mtu;
1267 enum ib_mig_state path_mig_state;
1272 int qp_access_flags;
1273 struct ib_qp_cap cap;
1274 struct rdma_ah_attr ah_attr;
1275 struct rdma_ah_attr alt_ah_attr;
1278 u8 en_sqd_async_notify;
1281 u8 max_dest_rd_atomic;
1290 struct net_device *xmit_slave;
1294 /* These are shared with userspace */
1295 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1296 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1297 IB_WR_SEND = IB_UVERBS_WR_SEND,
1298 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1299 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1300 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1301 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1302 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1303 IB_WR_LSO = IB_UVERBS_WR_TSO,
1304 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1305 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1306 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1307 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1308 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1309 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1310 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1312 /* These are kernel only and can not be issued by userspace */
1313 IB_WR_REG_MR = 0x20,
1314 IB_WR_REG_MR_INTEGRITY,
1316 /* reserve values for low level drivers' internal use.
1317 * These values will not be used at all in the ib core layer.
1319 IB_WR_RESERVED1 = 0xf0,
1331 enum ib_send_flags {
1333 IB_SEND_SIGNALED = (1<<1),
1334 IB_SEND_SOLICITED = (1<<2),
1335 IB_SEND_INLINE = (1<<3),
1336 IB_SEND_IP_CSUM = (1<<4),
1338 /* reserve bits 26-31 for low level drivers' internal use */
1339 IB_SEND_RESERVED_START = (1 << 26),
1340 IB_SEND_RESERVED_END = (1 << 31),
1350 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1354 struct ib_send_wr *next;
1357 struct ib_cqe *wr_cqe;
1359 struct ib_sge *sg_list;
1361 enum ib_wr_opcode opcode;
1365 u32 invalidate_rkey;
1370 struct ib_send_wr wr;
1375 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1377 return container_of(wr, struct ib_rdma_wr, wr);
1380 struct ib_atomic_wr {
1381 struct ib_send_wr wr;
1385 u64 compare_add_mask;
1390 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1392 return container_of(wr, struct ib_atomic_wr, wr);
1396 struct ib_send_wr wr;
1403 u16 pkey_index; /* valid for GSI only */
1404 u32 port_num; /* valid for DR SMPs on switch only */
1407 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1409 return container_of(wr, struct ib_ud_wr, wr);
1413 struct ib_send_wr wr;
1419 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1421 return container_of(wr, struct ib_reg_wr, wr);
1425 struct ib_recv_wr *next;
1428 struct ib_cqe *wr_cqe;
1430 struct ib_sge *sg_list;
1434 enum ib_access_flags {
1435 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1436 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1437 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1438 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1439 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1440 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1441 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1442 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1443 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1445 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1446 IB_ACCESS_SUPPORTED =
1447 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1451 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1452 * are hidden here instead of a uapi header!
1454 enum ib_mr_rereg_flags {
1455 IB_MR_REREG_TRANS = 1,
1456 IB_MR_REREG_PD = (1<<1),
1457 IB_MR_REREG_ACCESS = (1<<2),
1458 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1463 enum rdma_remove_reason {
1465 * Userspace requested uobject deletion or initial try
1466 * to remove uobject via cleanup. Call could fail
1468 RDMA_REMOVE_DESTROY,
1469 /* Context deletion. This call should delete the actual object itself */
1471 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1472 RDMA_REMOVE_DRIVER_REMOVE,
1473 /* uobj is being cleaned-up before being committed */
1475 /* The driver failed to destroy the uobject and is being disconnected */
1476 RDMA_REMOVE_DRIVER_FAILURE,
1479 struct ib_rdmacg_object {
1480 #ifdef CONFIG_CGROUP_RDMA
1481 struct rdma_cgroup *cg; /* owner rdma cgroup */
1485 struct ib_ucontext {
1486 struct ib_device *device;
1487 struct ib_uverbs_file *ufile;
1489 struct ib_rdmacg_object cg_obj;
1491 * Implementation details of the RDMA core, don't use in drivers:
1493 struct rdma_restrack_entry res;
1494 struct xarray mmap_xa;
1498 u64 user_handle; /* handle given to us by userspace */
1499 /* ufile & ucontext owning this object */
1500 struct ib_uverbs_file *ufile;
1501 /* FIXME, save memory: ufile->context == context */
1502 struct ib_ucontext *context; /* associated user context */
1503 void *object; /* containing object */
1504 struct list_head list; /* link to context's list */
1505 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1506 int id; /* index into kernel idr */
1508 atomic_t usecnt; /* protects exclusive access */
1509 struct rcu_head rcu; /* kfree_rcu() overhead */
1511 const struct uverbs_api_object *uapi_object;
1515 const void __user *inbuf;
1516 void __user *outbuf;
1524 struct ib_device *device;
1525 struct ib_uobject *uobject;
1526 atomic_t usecnt; /* count all resources */
1528 u32 unsafe_global_rkey;
1531 * Implementation details of the RDMA core, don't use in drivers:
1533 struct ib_mr *__internal_mr;
1534 struct rdma_restrack_entry res;
1538 struct ib_device *device;
1539 atomic_t usecnt; /* count all exposed resources */
1540 struct inode *inode;
1541 struct rw_semaphore tgt_qps_rwsem;
1542 struct xarray tgt_qps;
1546 struct ib_device *device;
1548 struct ib_uobject *uobject;
1549 const struct ib_gid_attr *sgid_attr;
1550 enum rdma_ah_attr_type type;
1553 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1555 enum ib_poll_context {
1556 IB_POLL_SOFTIRQ, /* poll from softirq context */
1557 IB_POLL_WORKQUEUE, /* poll from workqueue */
1558 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1559 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1561 IB_POLL_DIRECT, /* caller context, no hw completions */
1565 struct ib_device *device;
1566 struct ib_ucq_object *uobject;
1567 ib_comp_handler comp_handler;
1568 void (*event_handler)(struct ib_event *, void *);
1571 unsigned int cqe_used;
1572 atomic_t usecnt; /* count number of work queues */
1573 enum ib_poll_context poll_ctx;
1575 struct list_head pool_entry;
1577 struct irq_poll iop;
1578 struct work_struct work;
1580 struct workqueue_struct *comp_wq;
1583 /* updated only by trace points */
1587 unsigned int comp_vector;
1590 * Implementation details of the RDMA core, don't use in drivers:
1592 struct rdma_restrack_entry res;
1596 struct ib_device *device;
1598 struct ib_usrq_object *uobject;
1599 void (*event_handler)(struct ib_event *, void *);
1601 enum ib_srq_type srq_type;
1608 struct ib_xrcd *xrcd;
1615 * Implementation details of the RDMA core, don't use in drivers:
1617 struct rdma_restrack_entry res;
1620 enum ib_raw_packet_caps {
1621 /* Strip cvlan from incoming packet and report it in the matching work
1622 * completion is supported.
1624 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1625 /* Scatter FCS field of an incoming packet to host memory is supported.
1627 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1628 /* Checksum offloads are supported (for both send and receive). */
1629 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1630 /* When a packet is received for an RQ with no receive WQEs, the
1631 * packet processing is delayed.
1633 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1637 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1647 struct ib_device *device;
1648 struct ib_uwq_object *uobject;
1650 void (*event_handler)(struct ib_event *, void *);
1654 enum ib_wq_state state;
1655 enum ib_wq_type wq_type;
1660 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1661 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1662 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1663 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1664 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1667 struct ib_wq_init_attr {
1669 enum ib_wq_type wq_type;
1673 void (*event_handler)(struct ib_event *, void *);
1674 u32 create_flags; /* Use enum ib_wq_flags */
1677 enum ib_wq_attr_mask {
1678 IB_WQ_STATE = 1 << 0,
1679 IB_WQ_CUR_STATE = 1 << 1,
1680 IB_WQ_FLAGS = 1 << 2,
1684 enum ib_wq_state wq_state;
1685 enum ib_wq_state curr_wq_state;
1686 u32 flags; /* Use enum ib_wq_flags */
1687 u32 flags_mask; /* Use enum ib_wq_flags */
1690 struct ib_rwq_ind_table {
1691 struct ib_device *device;
1692 struct ib_uobject *uobject;
1695 u32 log_ind_tbl_size;
1696 struct ib_wq **ind_tbl;
1699 struct ib_rwq_ind_table_init_attr {
1700 u32 log_ind_tbl_size;
1701 /* Each entry is a pointer to Receive Work Queue */
1702 struct ib_wq **ind_tbl;
1705 enum port_pkey_state {
1706 IB_PORT_PKEY_NOT_VALID = 0,
1707 IB_PORT_PKEY_VALID = 1,
1708 IB_PORT_PKEY_LISTED = 2,
1711 struct ib_qp_security;
1713 struct ib_port_pkey {
1714 enum port_pkey_state state;
1717 struct list_head qp_list;
1718 struct list_head to_error_list;
1719 struct ib_qp_security *sec;
1722 struct ib_ports_pkeys {
1723 struct ib_port_pkey main;
1724 struct ib_port_pkey alt;
1727 struct ib_qp_security {
1729 struct ib_device *dev;
1730 /* Hold this mutex when changing port and pkey settings. */
1732 struct ib_ports_pkeys *ports_pkeys;
1733 /* A list of all open shared QP handles. Required to enforce security
1734 * properly for all users of a shared QP.
1736 struct list_head shared_qp_list;
1739 atomic_t error_list_count;
1740 struct completion error_complete;
1741 int error_comps_pending;
1745 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1746 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1749 struct ib_device *device;
1751 struct ib_cq *send_cq;
1752 struct ib_cq *recv_cq;
1755 struct list_head rdma_mrs;
1756 struct list_head sig_mrs;
1758 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1759 struct list_head xrcd_list;
1761 /* count times opened, mcast attaches, flow attaches */
1763 struct list_head open_list;
1764 struct ib_qp *real_qp;
1765 struct ib_uqp_object *uobject;
1766 void (*event_handler)(struct ib_event *, void *);
1768 /* sgid_attrs associated with the AV's */
1769 const struct ib_gid_attr *av_sgid_attr;
1770 const struct ib_gid_attr *alt_path_sgid_attr;
1774 enum ib_qp_type qp_type;
1775 struct ib_rwq_ind_table *rwq_ind_tbl;
1776 struct ib_qp_security *qp_sec;
1781 * Implementation details of the RDMA core, don't use in drivers:
1783 struct rdma_restrack_entry res;
1785 /* The counter the qp is bind to */
1786 struct rdma_counter *counter;
1790 struct ib_device *device;
1793 struct ib_uobject *uobject;
1798 struct ib_device *device;
1804 unsigned int page_size;
1805 enum ib_mr_type type;
1808 struct ib_uobject *uobject; /* user */
1809 struct list_head qp_entry; /* FR */
1813 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1815 * Implementation details of the RDMA core, don't use in drivers:
1817 struct rdma_restrack_entry res;
1821 struct ib_device *device;
1823 struct ib_uobject *uobject;
1825 enum ib_mw_type type;
1828 /* Supported steering options */
1829 enum ib_flow_attr_type {
1830 /* steering according to rule specifications */
1831 IB_FLOW_ATTR_NORMAL = 0x0,
1832 /* default unicast and multicast rule -
1833 * receive all Eth traffic which isn't steered to any QP
1835 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1836 /* default multicast rule -
1837 * receive all Eth multicast traffic which isn't steered to any QP
1839 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1840 /* sniffer rule - receive all port traffic */
1841 IB_FLOW_ATTR_SNIFFER = 0x3
1844 /* Supported steering header types */
1845 enum ib_flow_spec_type {
1847 IB_FLOW_SPEC_ETH = 0x20,
1848 IB_FLOW_SPEC_IB = 0x22,
1850 IB_FLOW_SPEC_IPV4 = 0x30,
1851 IB_FLOW_SPEC_IPV6 = 0x31,
1852 IB_FLOW_SPEC_ESP = 0x34,
1854 IB_FLOW_SPEC_TCP = 0x40,
1855 IB_FLOW_SPEC_UDP = 0x41,
1856 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1857 IB_FLOW_SPEC_GRE = 0x51,
1858 IB_FLOW_SPEC_MPLS = 0x60,
1859 IB_FLOW_SPEC_INNER = 0x100,
1861 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1862 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1863 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1864 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1866 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1867 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1869 enum ib_flow_flags {
1870 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1871 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1872 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1875 struct ib_flow_eth_filter {
1884 struct ib_flow_spec_eth {
1887 struct ib_flow_eth_filter val;
1888 struct ib_flow_eth_filter mask;
1891 struct ib_flow_ib_filter {
1898 struct ib_flow_spec_ib {
1901 struct ib_flow_ib_filter val;
1902 struct ib_flow_ib_filter mask;
1905 /* IPv4 header flags */
1906 enum ib_ipv4_flags {
1907 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1908 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1909 last have this flag set */
1912 struct ib_flow_ipv4_filter {
1923 struct ib_flow_spec_ipv4 {
1926 struct ib_flow_ipv4_filter val;
1927 struct ib_flow_ipv4_filter mask;
1930 struct ib_flow_ipv6_filter {
1941 struct ib_flow_spec_ipv6 {
1944 struct ib_flow_ipv6_filter val;
1945 struct ib_flow_ipv6_filter mask;
1948 struct ib_flow_tcp_udp_filter {
1955 struct ib_flow_spec_tcp_udp {
1958 struct ib_flow_tcp_udp_filter val;
1959 struct ib_flow_tcp_udp_filter mask;
1962 struct ib_flow_tunnel_filter {
1967 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1968 * the tunnel_id from val has the vni value
1970 struct ib_flow_spec_tunnel {
1973 struct ib_flow_tunnel_filter val;
1974 struct ib_flow_tunnel_filter mask;
1977 struct ib_flow_esp_filter {
1984 struct ib_flow_spec_esp {
1987 struct ib_flow_esp_filter val;
1988 struct ib_flow_esp_filter mask;
1991 struct ib_flow_gre_filter {
1992 __be16 c_ks_res0_ver;
1999 struct ib_flow_spec_gre {
2002 struct ib_flow_gre_filter val;
2003 struct ib_flow_gre_filter mask;
2006 struct ib_flow_mpls_filter {
2012 struct ib_flow_spec_mpls {
2015 struct ib_flow_mpls_filter val;
2016 struct ib_flow_mpls_filter mask;
2019 struct ib_flow_spec_action_tag {
2020 enum ib_flow_spec_type type;
2025 struct ib_flow_spec_action_drop {
2026 enum ib_flow_spec_type type;
2030 struct ib_flow_spec_action_handle {
2031 enum ib_flow_spec_type type;
2033 struct ib_flow_action *act;
2036 enum ib_counters_description {
2041 struct ib_flow_spec_action_count {
2042 enum ib_flow_spec_type type;
2044 struct ib_counters *counters;
2047 union ib_flow_spec {
2052 struct ib_flow_spec_eth eth;
2053 struct ib_flow_spec_ib ib;
2054 struct ib_flow_spec_ipv4 ipv4;
2055 struct ib_flow_spec_tcp_udp tcp_udp;
2056 struct ib_flow_spec_ipv6 ipv6;
2057 struct ib_flow_spec_tunnel tunnel;
2058 struct ib_flow_spec_esp esp;
2059 struct ib_flow_spec_gre gre;
2060 struct ib_flow_spec_mpls mpls;
2061 struct ib_flow_spec_action_tag flow_tag;
2062 struct ib_flow_spec_action_drop drop;
2063 struct ib_flow_spec_action_handle action;
2064 struct ib_flow_spec_action_count flow_count;
2067 struct ib_flow_attr {
2068 enum ib_flow_attr_type type;
2074 union ib_flow_spec flows[];
2079 struct ib_device *device;
2080 struct ib_uobject *uobject;
2083 enum ib_flow_action_type {
2084 IB_FLOW_ACTION_UNSPECIFIED,
2085 IB_FLOW_ACTION_ESP = 1,
2088 struct ib_flow_action_attrs_esp_keymats {
2089 enum ib_uverbs_flow_action_esp_keymat protocol;
2091 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2095 struct ib_flow_action_attrs_esp_replays {
2096 enum ib_uverbs_flow_action_esp_replay protocol;
2098 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2102 enum ib_flow_action_attrs_esp_flags {
2103 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2104 * This is done in order to share the same flags between user-space and
2105 * kernel and spare an unnecessary translation.
2109 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2110 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2113 struct ib_flow_spec_list {
2114 struct ib_flow_spec_list *next;
2115 union ib_flow_spec spec;
2118 struct ib_flow_action_attrs_esp {
2119 struct ib_flow_action_attrs_esp_keymats *keymat;
2120 struct ib_flow_action_attrs_esp_replays *replay;
2121 struct ib_flow_spec_list *encap;
2122 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2123 * Value of 0 is a valid value.
2129 /* Use enum ib_flow_action_attrs_esp_flags */
2131 u64 hard_limit_pkts;
2134 struct ib_flow_action {
2135 struct ib_device *device;
2136 struct ib_uobject *uobject;
2137 enum ib_flow_action_type type;
2144 enum ib_process_mad_flags {
2145 IB_MAD_IGNORE_MKEY = 1,
2146 IB_MAD_IGNORE_BKEY = 2,
2147 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2150 enum ib_mad_result {
2151 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2152 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2153 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2154 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2157 struct ib_port_cache {
2159 struct ib_pkey_cache *pkey;
2160 struct ib_gid_table *gid;
2162 enum ib_port_state port_state;
2165 struct ib_port_immutable {
2172 struct ib_port_data {
2173 struct ib_device *ib_dev;
2175 struct ib_port_immutable immutable;
2177 spinlock_t pkey_list_lock;
2178 struct list_head pkey_list;
2180 struct ib_port_cache cache;
2182 spinlock_t netdev_lock;
2183 struct net_device __rcu *netdev;
2184 struct hlist_node ndev_hash_link;
2185 struct rdma_port_counter port_counter;
2186 struct rdma_hw_stats *hw_stats;
2189 /* rdma netdev type - specifies protocol type */
2190 enum rdma_netdev_t {
2191 RDMA_NETDEV_OPA_VNIC,
2196 * struct rdma_netdev - rdma netdev
2197 * For cases where netstack interfacing is required.
2199 struct rdma_netdev {
2201 struct ib_device *hca;
2206 * cleanup function must be specified.
2207 * FIXME: This is only used for OPA_VNIC and that usage should be
2210 void (*free_rdma_netdev)(struct net_device *netdev);
2212 /* control functions */
2213 void (*set_id)(struct net_device *netdev, int id);
2215 int (*send)(struct net_device *dev, struct sk_buff *skb,
2216 struct ib_ah *address, u32 dqpn);
2218 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2219 union ib_gid *gid, u16 mlid,
2220 int set_qkey, u32 qkey);
2221 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2222 union ib_gid *gid, u16 mlid);
2224 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2227 struct rdma_netdev_alloc_params {
2233 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2234 struct net_device *netdev, void *param);
2237 struct ib_odp_counters {
2239 atomic64_t invalidations;
2240 atomic64_t prefetch;
2243 struct ib_counters {
2244 struct ib_device *device;
2245 struct ib_uobject *uobject;
2246 /* num of objects attached */
2250 struct ib_counters_read_attr {
2253 u32 flags; /* use enum ib_read_counters_flags */
2256 struct uverbs_attr_bundle;
2258 struct iw_cm_conn_param;
2260 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2261 .size_##ib_struct = \
2262 (sizeof(struct drv_struct) + \
2263 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2264 BUILD_BUG_ON_ZERO( \
2265 !__same_type(((struct drv_struct *)NULL)->member, \
2268 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2269 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2271 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2272 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2274 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2276 struct rdma_user_mmap_entry {
2278 struct ib_ucontext *ucontext;
2279 unsigned long start_pgoff;
2281 bool driver_removed;
2284 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2286 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2288 return (u64)entry->start_pgoff << PAGE_SHIFT;
2292 * struct ib_device_ops - InfiniBand device operations
2293 * This structure defines all the InfiniBand device operations, providers will
2294 * need to define the supported operations, otherwise they will be set to null.
2296 struct ib_device_ops {
2297 struct module *owner;
2298 enum rdma_driver_id driver_id;
2300 unsigned int uverbs_no_driver_id_binding:1;
2302 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2303 const struct ib_send_wr **bad_send_wr);
2304 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2305 const struct ib_recv_wr **bad_recv_wr);
2306 void (*drain_rq)(struct ib_qp *qp);
2307 void (*drain_sq)(struct ib_qp *qp);
2308 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2309 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2310 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2311 int (*post_srq_recv)(struct ib_srq *srq,
2312 const struct ib_recv_wr *recv_wr,
2313 const struct ib_recv_wr **bad_recv_wr);
2314 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2315 u32 port_num, const struct ib_wc *in_wc,
2316 const struct ib_grh *in_grh,
2317 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2318 size_t *out_mad_size, u16 *out_mad_pkey_index);
2319 int (*query_device)(struct ib_device *device,
2320 struct ib_device_attr *device_attr,
2321 struct ib_udata *udata);
2322 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2323 struct ib_device_modify *device_modify);
2324 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2325 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2327 int (*query_port)(struct ib_device *device, u32 port_num,
2328 struct ib_port_attr *port_attr);
2329 int (*modify_port)(struct ib_device *device, u32 port_num,
2330 int port_modify_mask,
2331 struct ib_port_modify *port_modify);
2333 * The following mandatory functions are used only at device
2334 * registration. Keep functions such as these at the end of this
2335 * structure to avoid cache line misses when accessing struct ib_device
2338 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2339 struct ib_port_immutable *immutable);
2340 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2343 * When calling get_netdev, the HW vendor's driver should return the
2344 * net device of device @device at port @port_num or NULL if such
2345 * a net device doesn't exist. The vendor driver should call dev_hold
2346 * on this net device. The HW vendor's device driver must guarantee
2347 * that this function returns NULL before the net device has finished
2348 * NETDEV_UNREGISTER state.
2350 struct net_device *(*get_netdev)(struct ib_device *device,
2353 * rdma netdev operation
2355 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2356 * must return -EOPNOTSUPP if it doesn't support the specified type.
2358 struct net_device *(*alloc_rdma_netdev)(
2359 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2360 const char *name, unsigned char name_assign_type,
2361 void (*setup)(struct net_device *));
2363 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2364 enum rdma_netdev_t type,
2365 struct rdma_netdev_alloc_params *params);
2367 * query_gid should be return GID value for @device, when @port_num
2368 * link layer is either IB or iWarp. It is no-op if @port_num port
2369 * is RoCE link layer.
2371 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2374 * When calling add_gid, the HW vendor's driver should add the gid
2375 * of device of port at gid index available at @attr. Meta-info of
2376 * that gid (for example, the network device related to this gid) is
2377 * available at @attr. @context allows the HW vendor driver to store
2378 * extra information together with a GID entry. The HW vendor driver may
2379 * allocate memory to contain this information and store it in @context
2380 * when a new GID entry is written to. Params are consistent until the
2381 * next call of add_gid or delete_gid. The function should return 0 on
2382 * success or error otherwise. The function could be called
2383 * concurrently for different ports. This function is only called when
2384 * roce_gid_table is used.
2386 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2388 * When calling del_gid, the HW vendor's driver should delete the
2389 * gid of device @device at gid index gid_index of port port_num
2390 * available in @attr.
2391 * Upon the deletion of a GID entry, the HW vendor must free any
2392 * allocated memory. The caller will clear @context afterwards.
2393 * This function is only called when roce_gid_table is used.
2395 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2396 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2398 int (*alloc_ucontext)(struct ib_ucontext *context,
2399 struct ib_udata *udata);
2400 void (*dealloc_ucontext)(struct ib_ucontext *context);
2401 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2403 * This will be called once refcount of an entry in mmap_xa reaches
2404 * zero. The type of the memory that was mapped may differ between
2405 * entries and is opaque to the rdma_user_mmap interface.
2406 * Therefore needs to be implemented by the driver in mmap_free.
2408 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2409 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2410 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2411 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2412 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2413 struct ib_udata *udata);
2414 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2415 struct ib_udata *udata);
2416 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2417 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2418 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2419 int (*create_srq)(struct ib_srq *srq,
2420 struct ib_srq_init_attr *srq_init_attr,
2421 struct ib_udata *udata);
2422 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2423 enum ib_srq_attr_mask srq_attr_mask,
2424 struct ib_udata *udata);
2425 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2426 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2427 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2428 struct ib_qp_init_attr *qp_init_attr,
2429 struct ib_udata *udata);
2430 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2431 int qp_attr_mask, struct ib_udata *udata);
2432 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2433 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2434 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2435 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2436 struct ib_udata *udata);
2437 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2438 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2439 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2440 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2441 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2442 u64 virt_addr, int mr_access_flags,
2443 struct ib_udata *udata);
2444 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2445 u64 length, u64 virt_addr, int fd,
2446 int mr_access_flags,
2447 struct ib_udata *udata);
2448 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2449 u64 length, u64 virt_addr,
2450 int mr_access_flags, struct ib_pd *pd,
2451 struct ib_udata *udata);
2452 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2453 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2455 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2456 u32 max_num_data_sg,
2457 u32 max_num_meta_sg);
2458 int (*advise_mr)(struct ib_pd *pd,
2459 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2460 struct ib_sge *sg_list, u32 num_sge,
2461 struct uverbs_attr_bundle *attrs);
2462 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2463 unsigned int *sg_offset);
2464 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2465 struct ib_mr_status *mr_status);
2466 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2467 int (*dealloc_mw)(struct ib_mw *mw);
2468 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2469 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2470 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2471 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2472 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2473 struct ib_flow_attr *flow_attr,
2474 struct ib_udata *udata);
2475 int (*destroy_flow)(struct ib_flow *flow_id);
2476 struct ib_flow_action *(*create_flow_action_esp)(
2477 struct ib_device *device,
2478 const struct ib_flow_action_attrs_esp *attr,
2479 struct uverbs_attr_bundle *attrs);
2480 int (*destroy_flow_action)(struct ib_flow_action *action);
2481 int (*modify_flow_action_esp)(
2482 struct ib_flow_action *action,
2483 const struct ib_flow_action_attrs_esp *attr,
2484 struct uverbs_attr_bundle *attrs);
2485 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2487 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2488 struct ifla_vf_info *ivf);
2489 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2490 struct ifla_vf_stats *stats);
2491 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2492 struct ifla_vf_guid *node_guid,
2493 struct ifla_vf_guid *port_guid);
2494 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2496 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2497 struct ib_wq_init_attr *init_attr,
2498 struct ib_udata *udata);
2499 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2500 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2501 u32 wq_attr_mask, struct ib_udata *udata);
2502 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2503 struct ib_rwq_ind_table_init_attr *init_attr,
2504 struct ib_udata *udata);
2505 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2506 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2507 struct ib_ucontext *context,
2508 struct ib_dm_alloc_attr *attr,
2509 struct uverbs_attr_bundle *attrs);
2510 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2511 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2512 struct ib_dm_mr_attr *attr,
2513 struct uverbs_attr_bundle *attrs);
2514 int (*create_counters)(struct ib_counters *counters,
2515 struct uverbs_attr_bundle *attrs);
2516 int (*destroy_counters)(struct ib_counters *counters);
2517 int (*read_counters)(struct ib_counters *counters,
2518 struct ib_counters_read_attr *counters_read_attr,
2519 struct uverbs_attr_bundle *attrs);
2520 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2521 int data_sg_nents, unsigned int *data_sg_offset,
2522 struct scatterlist *meta_sg, int meta_sg_nents,
2523 unsigned int *meta_sg_offset);
2526 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2527 * driver initialized data. The struct is kfree()'ed by the sysfs
2528 * core when the device is removed. A lifespan of -1 in the return
2529 * struct tells the core to set a default lifespan.
2531 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2534 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2535 * @index - The index in the value array we wish to have updated, or
2536 * num_counters if we want all stats updated
2538 * < 0 - Error, no counters updated
2539 * index - Updated the single counter pointed to by index
2540 * num_counters - Updated all counters (will reset the timestamp
2541 * and prevent further calls for lifespan milliseconds)
2542 * Drivers are allowed to update all counters in leiu of just the
2543 * one given in index at their option
2545 int (*get_hw_stats)(struct ib_device *device,
2546 struct rdma_hw_stats *stats, u32 port, int index);
2548 * This function is called once for each port when a ib device is
2551 int (*init_port)(struct ib_device *device, u32 port_num,
2552 struct kobject *port_sysfs);
2554 * Allows rdma drivers to add their own restrack attributes.
2556 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2557 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2558 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2559 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2560 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2561 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2562 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2564 /* Device lifecycle callbacks */
2566 * Called after the device becomes registered, before clients are
2569 int (*enable_driver)(struct ib_device *dev);
2571 * This is called as part of ib_dealloc_device().
2573 void (*dealloc_driver)(struct ib_device *dev);
2575 /* iWarp CM callbacks */
2576 void (*iw_add_ref)(struct ib_qp *qp);
2577 void (*iw_rem_ref)(struct ib_qp *qp);
2578 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2579 int (*iw_connect)(struct iw_cm_id *cm_id,
2580 struct iw_cm_conn_param *conn_param);
2581 int (*iw_accept)(struct iw_cm_id *cm_id,
2582 struct iw_cm_conn_param *conn_param);
2583 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2585 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2586 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2588 * counter_bind_qp - Bind a QP to a counter.
2589 * @counter - The counter to be bound. If counter->id is zero then
2590 * the driver needs to allocate a new counter and set counter->id
2592 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2594 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2595 * counter and bind it onto the default one
2597 int (*counter_unbind_qp)(struct ib_qp *qp);
2599 * counter_dealloc -De-allocate the hw counter
2601 int (*counter_dealloc)(struct rdma_counter *counter);
2603 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2604 * the driver initialized data.
2606 struct rdma_hw_stats *(*counter_alloc_stats)(
2607 struct rdma_counter *counter);
2609 * counter_update_stats - Query the stats value of this counter
2611 int (*counter_update_stats)(struct rdma_counter *counter);
2614 * Allows rdma drivers to add their own restrack attributes
2615 * dumped via 'rdma stat' iproute2 command.
2617 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2619 /* query driver for its ucontext properties */
2620 int (*query_ucontext)(struct ib_ucontext *context,
2621 struct uverbs_attr_bundle *attrs);
2623 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2624 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2625 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2626 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2627 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2628 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2629 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2630 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2631 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2634 struct ib_core_device {
2635 /* device must be the first element in structure until,
2636 * union of ib_core_device and device exists in ib_device.
2639 possible_net_t rdma_net;
2640 struct kobject *ports_kobj;
2641 struct list_head port_list;
2642 struct ib_device *owner; /* reach back to owner ib_device */
2645 struct rdma_restrack_root;
2647 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2648 struct device *dma_device;
2649 struct ib_device_ops ops;
2650 char name[IB_DEVICE_NAME_MAX];
2651 struct rcu_head rcu_head;
2653 struct list_head event_handler_list;
2654 /* Protects event_handler_list */
2655 struct rw_semaphore event_handler_rwsem;
2657 /* Protects QP's event_handler calls and open_qp list */
2658 spinlock_t qp_open_list_lock;
2660 struct rw_semaphore client_data_rwsem;
2661 struct xarray client_data;
2662 struct mutex unregistration_lock;
2664 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2665 rwlock_t cache_lock;
2667 * port_data is indexed by port number
2669 struct ib_port_data *port_data;
2671 int num_comp_vectors;
2675 struct ib_core_device coredev;
2678 /* First group for device attributes,
2679 * Second group for driver provided attributes (optional).
2680 * It is NULL terminated array.
2682 const struct attribute_group *groups[3];
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 attribute_group *hw_stats_ag;
2698 struct rdma_hw_stats *hw_stats;
2700 #ifdef CONFIG_CGROUP_RDMA
2701 struct rdmacg_device cg_device;
2706 spinlock_t cq_pools_lock;
2707 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2709 struct rdma_restrack_root *res;
2711 const struct uapi_definition *driver_def;
2714 * Positive refcount indicates that the device is currently
2715 * registered and cannot be unregistered.
2717 refcount_t refcount;
2718 struct completion unreg_completion;
2719 struct work_struct unregistration_work;
2721 const struct rdma_link_ops *link_ops;
2723 /* Protects compat_devs xarray modifications */
2724 struct mutex compat_devs_mutex;
2725 /* Maintains compat devices for each net namespace */
2726 struct xarray compat_devs;
2728 /* Used by iWarp CM */
2729 char iw_ifname[IFNAMSIZ];
2730 u32 iw_driver_flags;
2734 struct ib_client_nl_info;
2737 int (*add)(struct ib_device *ibdev);
2738 void (*remove)(struct ib_device *, void *client_data);
2739 void (*rename)(struct ib_device *dev, void *client_data);
2740 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2741 struct ib_client_nl_info *res);
2742 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2744 /* Returns the net_dev belonging to this ib_client and matching the
2746 * @dev: An RDMA device that the net_dev use for communication.
2747 * @port: A physical port number on the RDMA device.
2748 * @pkey: P_Key that the net_dev uses if applicable.
2749 * @gid: A GID that the net_dev uses to communicate.
2750 * @addr: An IP address the net_dev is configured with.
2751 * @client_data: The device's client data set by ib_set_client_data().
2753 * An ib_client that implements a net_dev on top of RDMA devices
2754 * (such as IP over IB) should implement this callback, allowing the
2755 * rdma_cm module to find the right net_dev for a given request.
2757 * The caller is responsible for calling dev_put on the returned
2759 struct net_device *(*get_net_dev_by_params)(
2760 struct ib_device *dev,
2763 const union ib_gid *gid,
2764 const struct sockaddr *addr,
2768 struct completion uses_zero;
2771 /* kverbs are not required by the client */
2776 * IB block DMA iterator
2778 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2779 * to a HW supported page size.
2781 struct ib_block_iter {
2782 /* internal states */
2783 struct scatterlist *__sg; /* sg holding the current aligned block */
2784 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2785 unsigned int __sg_nents; /* number of SG entries */
2786 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2787 unsigned int __pg_bit; /* alignment of current block */
2790 struct ib_device *_ib_alloc_device(size_t size);
2791 #define ib_alloc_device(drv_struct, member) \
2792 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2793 BUILD_BUG_ON_ZERO(offsetof( \
2794 struct drv_struct, member))), \
2795 struct drv_struct, member)
2797 void ib_dealloc_device(struct ib_device *device);
2799 void ib_get_device_fw_str(struct ib_device *device, char *str);
2801 int ib_register_device(struct ib_device *device, const char *name,
2802 struct device *dma_device);
2803 void ib_unregister_device(struct ib_device *device);
2804 void ib_unregister_driver(enum rdma_driver_id driver_id);
2805 void ib_unregister_device_and_put(struct ib_device *device);
2806 void ib_unregister_device_queued(struct ib_device *ib_dev);
2808 int ib_register_client (struct ib_client *client);
2809 void ib_unregister_client(struct ib_client *client);
2811 void __rdma_block_iter_start(struct ib_block_iter *biter,
2812 struct scatterlist *sglist,
2814 unsigned long pgsz);
2815 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2818 * rdma_block_iter_dma_address - get the aligned dma address of the current
2819 * block held by the block iterator.
2820 * @biter: block iterator holding the memory block
2822 static inline dma_addr_t
2823 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2825 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2829 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2830 * @sglist: sglist to iterate over
2831 * @biter: block iterator holding the memory block
2832 * @nents: maximum number of sg entries to iterate over
2833 * @pgsz: best HW supported page size to use
2835 * Callers may use rdma_block_iter_dma_address() to get each
2836 * blocks aligned DMA address.
2838 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2839 for (__rdma_block_iter_start(biter, sglist, nents, \
2841 __rdma_block_iter_next(biter);)
2844 * ib_get_client_data - Get IB client context
2845 * @device:Device to get context for
2846 * @client:Client to get context for
2848 * ib_get_client_data() returns the client context data set with
2849 * ib_set_client_data(). This can only be called while the client is
2850 * registered to the device, once the ib_client remove() callback returns this
2853 static inline void *ib_get_client_data(struct ib_device *device,
2854 struct ib_client *client)
2856 return xa_load(&device->client_data, client->client_id);
2858 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2860 void ib_set_device_ops(struct ib_device *device,
2861 const struct ib_device_ops *ops);
2863 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2864 unsigned long pfn, unsigned long size, pgprot_t prot,
2865 struct rdma_user_mmap_entry *entry);
2866 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2867 struct rdma_user_mmap_entry *entry,
2869 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2870 struct rdma_user_mmap_entry *entry,
2871 size_t length, u32 min_pgoff,
2874 struct rdma_user_mmap_entry *
2875 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2876 unsigned long pgoff);
2877 struct rdma_user_mmap_entry *
2878 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2879 struct vm_area_struct *vma);
2880 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2882 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2884 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2886 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2889 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2891 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2894 static inline bool ib_is_buffer_cleared(const void __user *p,
2900 if (len > USHRT_MAX)
2903 buf = memdup_user(p, len);
2907 ret = !memchr_inv(buf, 0, len);
2912 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2916 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2920 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2921 * contains all required attributes and no attributes not allowed for
2922 * the given QP state transition.
2923 * @cur_state: Current QP state
2924 * @next_state: Next QP state
2926 * @mask: Mask of supplied QP attributes
2928 * This function is a helper function that a low-level driver's
2929 * modify_qp method can use to validate the consumer's input. It
2930 * checks that cur_state and next_state are valid QP states, that a
2931 * transition from cur_state to next_state is allowed by the IB spec,
2932 * and that the attribute mask supplied is allowed for the transition.
2934 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2935 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2937 void ib_register_event_handler(struct ib_event_handler *event_handler);
2938 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2939 void ib_dispatch_event(const struct ib_event *event);
2941 int ib_query_port(struct ib_device *device,
2942 u32 port_num, struct ib_port_attr *port_attr);
2944 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2948 * rdma_cap_ib_switch - Check if the device is IB switch
2949 * @device: Device to check
2951 * Device driver is responsible for setting is_switch bit on
2952 * in ib_device structure at init time.
2954 * Return: true if the device is IB switch.
2956 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2958 return device->is_switch;
2962 * rdma_start_port - Return the first valid port number for the device
2965 * @device: Device to be checked
2967 * Return start port number
2969 static inline u32 rdma_start_port(const struct ib_device *device)
2971 return rdma_cap_ib_switch(device) ? 0 : 1;
2975 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2976 * @device - The struct ib_device * to iterate over
2977 * @iter - The unsigned int to store the port number
2979 #define rdma_for_each_port(device, iter) \
2980 for (iter = rdma_start_port(device + \
2981 BUILD_BUG_ON_ZERO(!__same_type(u32, \
2983 iter <= rdma_end_port(device); iter++)
2986 * rdma_end_port - Return the last valid port number for the device
2989 * @device: Device to be checked
2991 * Return last port number
2993 static inline u32 rdma_end_port(const struct ib_device *device)
2995 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2998 static inline int rdma_is_port_valid(const struct ib_device *device,
3001 return (port >= rdma_start_port(device) &&
3002 port <= rdma_end_port(device));
3005 static inline bool rdma_is_grh_required(const struct ib_device *device,
3008 return device->port_data[port_num].immutable.core_cap_flags &
3009 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3012 static inline bool rdma_protocol_ib(const struct ib_device *device,
3015 return device->port_data[port_num].immutable.core_cap_flags &
3016 RDMA_CORE_CAP_PROT_IB;
3019 static inline bool rdma_protocol_roce(const struct ib_device *device,
3022 return device->port_data[port_num].immutable.core_cap_flags &
3023 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3026 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3029 return device->port_data[port_num].immutable.core_cap_flags &
3030 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3033 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3036 return device->port_data[port_num].immutable.core_cap_flags &
3037 RDMA_CORE_CAP_PROT_ROCE;
3040 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3043 return device->port_data[port_num].immutable.core_cap_flags &
3044 RDMA_CORE_CAP_PROT_IWARP;
3047 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3050 return rdma_protocol_ib(device, port_num) ||
3051 rdma_protocol_roce(device, port_num);
3054 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3057 return device->port_data[port_num].immutable.core_cap_flags &
3058 RDMA_CORE_CAP_PROT_RAW_PACKET;
3061 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3064 return device->port_data[port_num].immutable.core_cap_flags &
3065 RDMA_CORE_CAP_PROT_USNIC;
3069 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3070 * Management Datagrams.
3071 * @device: Device to check
3072 * @port_num: Port number to check
3074 * Management Datagrams (MAD) are a required part of the InfiniBand
3075 * specification and are supported on all InfiniBand devices. A slightly
3076 * extended version are also supported on OPA interfaces.
3078 * Return: true if the port supports sending/receiving of MAD packets.
3080 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3082 return device->port_data[port_num].immutable.core_cap_flags &
3083 RDMA_CORE_CAP_IB_MAD;
3087 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3088 * Management Datagrams.
3089 * @device: Device to check
3090 * @port_num: Port number to check
3092 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3093 * datagrams with their own versions. These OPA MADs share many but not all of
3094 * the characteristics of InfiniBand MADs.
3096 * OPA MADs differ in the following ways:
3098 * 1) MADs are variable size up to 2K
3099 * IBTA defined MADs remain fixed at 256 bytes
3100 * 2) OPA SMPs must carry valid PKeys
3101 * 3) OPA SMP packets are a different format
3103 * Return: true if the port supports OPA MAD packet formats.
3105 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3107 return device->port_data[port_num].immutable.core_cap_flags &
3108 RDMA_CORE_CAP_OPA_MAD;
3112 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3113 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3114 * @device: Device to check
3115 * @port_num: Port number to check
3117 * Each InfiniBand node is required to provide a Subnet Management Agent
3118 * that the subnet manager can access. Prior to the fabric being fully
3119 * configured by the subnet manager, the SMA is accessed via a well known
3120 * interface called the Subnet Management Interface (SMI). This interface
3121 * uses directed route packets to communicate with the SM to get around the
3122 * chicken and egg problem of the SM needing to know what's on the fabric
3123 * in order to configure the fabric, and needing to configure the fabric in
3124 * order to send packets to the devices on the fabric. These directed
3125 * route packets do not need the fabric fully configured in order to reach
3126 * their destination. The SMI is the only method allowed to send
3127 * directed route packets on an InfiniBand fabric.
3129 * Return: true if the port provides an SMI.
3131 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3133 return device->port_data[port_num].immutable.core_cap_flags &
3134 RDMA_CORE_CAP_IB_SMI;
3138 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3139 * Communication Manager.
3140 * @device: Device to check
3141 * @port_num: Port number to check
3143 * The InfiniBand Communication Manager is one of many pre-defined General
3144 * Service Agents (GSA) that are accessed via the General Service
3145 * Interface (GSI). It's role is to facilitate establishment of connections
3146 * between nodes as well as other management related tasks for established
3149 * Return: true if the port supports an IB CM (this does not guarantee that
3150 * a CM is actually running however).
3152 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3154 return device->port_data[port_num].immutable.core_cap_flags &
3155 RDMA_CORE_CAP_IB_CM;
3159 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3160 * Communication Manager.
3161 * @device: Device to check
3162 * @port_num: Port number to check
3164 * Similar to above, but specific to iWARP connections which have a different
3165 * managment protocol than InfiniBand.
3167 * Return: true if the port supports an iWARP CM (this does not guarantee that
3168 * a CM is actually running however).
3170 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3172 return device->port_data[port_num].immutable.core_cap_flags &
3173 RDMA_CORE_CAP_IW_CM;
3177 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3178 * Subnet Administration.
3179 * @device: Device to check
3180 * @port_num: Port number to check
3182 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3183 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3184 * fabrics, devices should resolve routes to other hosts by contacting the
3185 * SA to query the proper route.
3187 * Return: true if the port should act as a client to the fabric Subnet
3188 * Administration interface. This does not imply that the SA service is
3191 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3193 return device->port_data[port_num].immutable.core_cap_flags &
3194 RDMA_CORE_CAP_IB_SA;
3198 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3200 * @device: Device to check
3201 * @port_num: Port number to check
3203 * InfiniBand multicast registration is more complex than normal IPv4 or
3204 * IPv6 multicast registration. Each Host Channel Adapter must register
3205 * with the Subnet Manager when it wishes to join a multicast group. It
3206 * should do so only once regardless of how many queue pairs it subscribes
3207 * to this group. And it should leave the group only after all queue pairs
3208 * attached to the group have been detached.
3210 * Return: true if the port must undertake the additional adminstrative
3211 * overhead of registering/unregistering with the SM and tracking of the
3212 * total number of queue pairs attached to the multicast group.
3214 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3217 return rdma_cap_ib_sa(device, port_num);
3221 * rdma_cap_af_ib - Check if the port of device has the capability
3222 * Native Infiniband Address.
3223 * @device: Device to check
3224 * @port_num: Port number to check
3226 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3227 * GID. RoCE uses a different mechanism, but still generates a GID via
3228 * a prescribed mechanism and port specific data.
3230 * Return: true if the port uses a GID address to identify devices on the
3233 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3235 return device->port_data[port_num].immutable.core_cap_flags &
3236 RDMA_CORE_CAP_AF_IB;
3240 * rdma_cap_eth_ah - Check if the port of device has the capability
3241 * Ethernet Address Handle.
3242 * @device: Device to check
3243 * @port_num: Port number to check
3245 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3246 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3247 * port. Normally, packet headers are generated by the sending host
3248 * adapter, but when sending connectionless datagrams, we must manually
3249 * inject the proper headers for the fabric we are communicating over.
3251 * Return: true if we are running as a RoCE port and must force the
3252 * addition of a Global Route Header built from our Ethernet Address
3253 * Handle into our header list for connectionless packets.
3255 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3257 return device->port_data[port_num].immutable.core_cap_flags &
3258 RDMA_CORE_CAP_ETH_AH;
3262 * rdma_cap_opa_ah - Check if the port of device supports
3263 * OPA Address handles
3264 * @device: Device to check
3265 * @port_num: Port number to check
3267 * Return: true if we are running on an OPA device which supports
3268 * the extended OPA addressing.
3270 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3272 return (device->port_data[port_num].immutable.core_cap_flags &
3273 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3277 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3280 * @port_num: Port number
3282 * This MAD size includes the MAD headers and MAD payload. No other headers
3285 * Return the max MAD size required by the Port. Will return 0 if the port
3286 * does not support MADs
3288 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3291 return device->port_data[port_num].immutable.max_mad_size;
3295 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3296 * @device: Device to check
3297 * @port_num: Port number to check
3299 * RoCE GID table mechanism manages the various GIDs for a device.
3301 * NOTE: if allocating the port's GID table has failed, this call will still
3302 * return true, but any RoCE GID table API will fail.
3304 * Return: true if the port uses RoCE GID table mechanism in order to manage
3307 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3310 return rdma_protocol_roce(device, port_num) &&
3311 device->ops.add_gid && device->ops.del_gid;
3315 * Check if the device supports READ W/ INVALIDATE.
3317 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3320 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3321 * has support for it yet.
3323 return rdma_protocol_iwarp(dev, port_num);
3327 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3329 * @port_num: 1 based Port number
3331 * Return true if port is an Intel OPA port , false if not
3333 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3336 return (device->port_data[port_num].immutable.core_cap_flags &
3337 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3341 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3343 * @port_num: Port number
3344 * @mtu: enum value of MTU
3346 * Return the MTU size supported by the port as an integer value. Will return
3347 * -1 if enum value of mtu is not supported.
3349 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3352 if (rdma_core_cap_opa_port(device, port))
3353 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3355 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3359 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3361 * @port_num: Port number
3362 * @attr: port attribute
3364 * Return the MTU size supported by the port as an integer value.
3366 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3367 struct ib_port_attr *attr)
3369 if (rdma_core_cap_opa_port(device, port))
3370 return attr->phys_mtu;
3372 return ib_mtu_enum_to_int(attr->max_mtu);
3375 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3377 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3378 struct ifla_vf_info *info);
3379 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3380 struct ifla_vf_stats *stats);
3381 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3382 struct ifla_vf_guid *node_guid,
3383 struct ifla_vf_guid *port_guid);
3384 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3387 int ib_query_pkey(struct ib_device *device,
3388 u32 port_num, u16 index, u16 *pkey);
3390 int ib_modify_device(struct ib_device *device,
3391 int device_modify_mask,
3392 struct ib_device_modify *device_modify);
3394 int ib_modify_port(struct ib_device *device,
3395 u32 port_num, int port_modify_mask,
3396 struct ib_port_modify *port_modify);
3398 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3399 u32 *port_num, u16 *index);
3401 int ib_find_pkey(struct ib_device *device,
3402 u32 port_num, u16 pkey, u16 *index);
3406 * Create a memory registration for all memory in the system and place
3407 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3408 * ULPs to avoid the overhead of dynamic MRs.
3410 * This flag is generally considered unsafe and must only be used in
3411 * extremly trusted environments. Every use of it will log a warning
3412 * in the kernel log.
3414 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3417 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3418 const char *caller);
3421 * ib_alloc_pd - Allocates an unused protection domain.
3422 * @device: The device on which to allocate the protection domain.
3423 * @flags: protection domain flags
3425 * A protection domain object provides an association between QPs, shared
3426 * receive queues, address handles, memory regions, and memory windows.
3428 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3429 * memory operations.
3431 #define ib_alloc_pd(device, flags) \
3432 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3434 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3437 * ib_dealloc_pd - Deallocate kernel PD
3438 * @pd: The protection domain
3440 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3442 static inline void ib_dealloc_pd(struct ib_pd *pd)
3444 int ret = ib_dealloc_pd_user(pd, NULL);
3446 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3449 enum rdma_create_ah_flags {
3450 /* In a sleepable context */
3451 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3455 * rdma_create_ah - Creates an address handle for the given address vector.
3456 * @pd: The protection domain associated with the address handle.
3457 * @ah_attr: The attributes of the address vector.
3458 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3460 * The address handle is used to reference a local or global destination
3461 * in all UD QP post sends.
3463 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3467 * rdma_create_user_ah - Creates an address handle for the given address vector.
3468 * It resolves destination mac address for ah attribute of RoCE type.
3469 * @pd: The protection domain associated with the address handle.
3470 * @ah_attr: The attributes of the address vector.
3471 * @udata: pointer to user's input output buffer information need by
3474 * It returns 0 on success and returns appropriate error code on error.
3475 * The address handle is used to reference a local or global destination
3476 * in all UD QP post sends.
3478 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3479 struct rdma_ah_attr *ah_attr,
3480 struct ib_udata *udata);
3482 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3484 * @hdr: the L3 header to parse
3485 * @net_type: type of header to parse
3486 * @sgid: place to store source gid
3487 * @dgid: place to store destination gid
3489 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3490 enum rdma_network_type net_type,
3491 union ib_gid *sgid, union ib_gid *dgid);
3494 * ib_get_rdma_header_version - Get the header version
3495 * @hdr: the L3 header to parse
3497 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3500 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3502 * @device: Device on which the received message arrived.
3503 * @port_num: Port on which the received message arrived.
3504 * @wc: Work completion associated with the received message.
3505 * @grh: References the received global route header. This parameter is
3506 * ignored unless the work completion indicates that the GRH is valid.
3507 * @ah_attr: Returned attributes that can be used when creating an address
3508 * handle for replying to the message.
3509 * When ib_init_ah_attr_from_wc() returns success,
3510 * (a) for IB link layer it optionally contains a reference to SGID attribute
3511 * when GRH is present for IB link layer.
3512 * (b) for RoCE link layer it contains a reference to SGID attribute.
3513 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3514 * attributes which are initialized using ib_init_ah_attr_from_wc().
3517 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3518 const struct ib_wc *wc, const struct ib_grh *grh,
3519 struct rdma_ah_attr *ah_attr);
3522 * ib_create_ah_from_wc - Creates an address handle associated with the
3523 * sender of the specified work completion.
3524 * @pd: The protection domain associated with the address handle.
3525 * @wc: Work completion information associated with a received message.
3526 * @grh: References the received global route header. This parameter is
3527 * ignored unless the work completion indicates that the GRH is valid.
3528 * @port_num: The outbound port number to associate with the address.
3530 * The address handle is used to reference a local or global destination
3531 * in all UD QP post sends.
3533 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3534 const struct ib_grh *grh, u32 port_num);
3537 * rdma_modify_ah - Modifies the address vector associated with an address
3539 * @ah: The address handle to modify.
3540 * @ah_attr: The new address vector attributes to associate with the
3543 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3546 * rdma_query_ah - Queries the address vector associated with an address
3548 * @ah: The address handle to query.
3549 * @ah_attr: The address vector attributes associated with the address
3552 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3554 enum rdma_destroy_ah_flags {
3555 /* In a sleepable context */
3556 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3560 * rdma_destroy_ah_user - Destroys an address handle.
3561 * @ah: The address handle to destroy.
3562 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3563 * @udata: Valid user data or NULL for kernel objects
3565 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3568 * rdma_destroy_ah - Destroys an kernel address handle.
3569 * @ah: The address handle to destroy.
3570 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3572 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3574 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3576 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3578 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3581 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3582 struct ib_srq_init_attr *srq_init_attr,
3583 struct ib_usrq_object *uobject,
3584 struct ib_udata *udata);
3585 static inline struct ib_srq *
3586 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3588 if (!pd->device->ops.create_srq)
3589 return ERR_PTR(-EOPNOTSUPP);
3591 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3595 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3596 * @srq: The SRQ to modify.
3597 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3598 * the current values of selected SRQ attributes are returned.
3599 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3600 * are being modified.
3602 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3603 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3604 * the number of receives queued drops below the limit.
3606 int ib_modify_srq(struct ib_srq *srq,
3607 struct ib_srq_attr *srq_attr,
3608 enum ib_srq_attr_mask srq_attr_mask);
3611 * ib_query_srq - Returns the attribute list and current values for the
3613 * @srq: The SRQ to query.
3614 * @srq_attr: The attributes of the specified SRQ.
3616 int ib_query_srq(struct ib_srq *srq,
3617 struct ib_srq_attr *srq_attr);
3620 * ib_destroy_srq_user - Destroys the specified SRQ.
3621 * @srq: The SRQ to destroy.
3622 * @udata: Valid user data or NULL for kernel objects
3624 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3627 * ib_destroy_srq - Destroys the specified kernel SRQ.
3628 * @srq: The SRQ to destroy.
3630 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3632 static inline void ib_destroy_srq(struct ib_srq *srq)
3634 int ret = ib_destroy_srq_user(srq, NULL);
3636 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3640 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3641 * @srq: The SRQ to post the work request on.
3642 * @recv_wr: A list of work requests to post on the receive queue.
3643 * @bad_recv_wr: On an immediate failure, this parameter will reference
3644 * the work request that failed to be posted on the QP.
3646 static inline int ib_post_srq_recv(struct ib_srq *srq,
3647 const struct ib_recv_wr *recv_wr,
3648 const struct ib_recv_wr **bad_recv_wr)
3650 const struct ib_recv_wr *dummy;
3652 return srq->device->ops.post_srq_recv(srq, recv_wr,
3653 bad_recv_wr ? : &dummy);
3656 struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
3657 struct ib_qp_init_attr *qp_init_attr,
3658 const char *caller);
3659 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3660 struct ib_qp_init_attr *init_attr)
3662 return ib_create_named_qp(pd, init_attr, KBUILD_MODNAME);
3666 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3667 * @qp: The QP to modify.
3668 * @attr: On input, specifies the QP attributes to modify. On output,
3669 * the current values of selected QP attributes are returned.
3670 * @attr_mask: A bit-mask used to specify which attributes of the QP
3671 * are being modified.
3672 * @udata: pointer to user's input output buffer information
3673 * are being modified.
3674 * It returns 0 on success and returns appropriate error code on error.
3676 int ib_modify_qp_with_udata(struct ib_qp *qp,
3677 struct ib_qp_attr *attr,
3679 struct ib_udata *udata);
3682 * ib_modify_qp - Modifies the attributes for the specified QP and then
3683 * transitions the QP to the given state.
3684 * @qp: The QP to modify.
3685 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3686 * the current values of selected QP attributes are returned.
3687 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3688 * are being modified.
3690 int ib_modify_qp(struct ib_qp *qp,
3691 struct ib_qp_attr *qp_attr,
3695 * ib_query_qp - Returns the attribute list and current values for the
3697 * @qp: The QP to query.
3698 * @qp_attr: The attributes of the specified QP.
3699 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3700 * @qp_init_attr: Additional attributes of the selected QP.
3702 * The qp_attr_mask may be used to limit the query to gathering only the
3703 * selected attributes.
3705 int ib_query_qp(struct ib_qp *qp,
3706 struct ib_qp_attr *qp_attr,
3708 struct ib_qp_init_attr *qp_init_attr);
3711 * ib_destroy_qp - Destroys the specified QP.
3712 * @qp: The QP to destroy.
3713 * @udata: Valid udata or NULL for kernel objects
3715 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3718 * ib_destroy_qp - Destroys the specified kernel QP.
3719 * @qp: The QP to destroy.
3721 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3723 static inline int ib_destroy_qp(struct ib_qp *qp)
3725 return ib_destroy_qp_user(qp, NULL);
3729 * ib_open_qp - Obtain a reference to an existing sharable QP.
3730 * @xrcd - XRC domain
3731 * @qp_open_attr: Attributes identifying the QP to open.
3733 * Returns a reference to a sharable QP.
3735 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3736 struct ib_qp_open_attr *qp_open_attr);
3739 * ib_close_qp - Release an external reference to a QP.
3740 * @qp: The QP handle to release
3742 * The opened QP handle is released by the caller. The underlying
3743 * shared QP is not destroyed until all internal references are released.
3745 int ib_close_qp(struct ib_qp *qp);
3748 * ib_post_send - Posts a list of work requests to the send queue of
3750 * @qp: The QP to post the work request on.
3751 * @send_wr: A list of work requests to post on the send queue.
3752 * @bad_send_wr: On an immediate failure, this parameter will reference
3753 * the work request that failed to be posted on the QP.
3755 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3756 * error is returned, the QP state shall not be affected,
3757 * ib_post_send() will return an immediate error after queueing any
3758 * earlier work requests in the list.
3760 static inline int ib_post_send(struct ib_qp *qp,
3761 const struct ib_send_wr *send_wr,
3762 const struct ib_send_wr **bad_send_wr)
3764 const struct ib_send_wr *dummy;
3766 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3770 * ib_post_recv - Posts a list of work requests to the receive queue of
3772 * @qp: The QP to post the work request on.
3773 * @recv_wr: A list of work requests to post on the receive queue.
3774 * @bad_recv_wr: On an immediate failure, this parameter will reference
3775 * the work request that failed to be posted on the QP.
3777 static inline int ib_post_recv(struct ib_qp *qp,
3778 const struct ib_recv_wr *recv_wr,
3779 const struct ib_recv_wr **bad_recv_wr)
3781 const struct ib_recv_wr *dummy;
3783 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3786 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3787 int comp_vector, enum ib_poll_context poll_ctx,
3788 const char *caller);
3789 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3790 int nr_cqe, int comp_vector,
3791 enum ib_poll_context poll_ctx)
3793 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3797 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3798 int nr_cqe, enum ib_poll_context poll_ctx,
3799 const char *caller);
3802 * ib_alloc_cq_any: Allocate kernel CQ
3803 * @dev: The IB device
3804 * @private: Private data attached to the CQE
3805 * @nr_cqe: Number of CQEs in the CQ
3806 * @poll_ctx: Context used for polling the CQ
3808 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3809 void *private, int nr_cqe,
3810 enum ib_poll_context poll_ctx)
3812 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3816 void ib_free_cq(struct ib_cq *cq);
3817 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3820 * ib_create_cq - Creates a CQ on the specified device.
3821 * @device: The device on which to create the CQ.
3822 * @comp_handler: A user-specified callback that is invoked when a
3823 * completion event occurs on the CQ.
3824 * @event_handler: A user-specified callback that is invoked when an
3825 * asynchronous event not associated with a completion occurs on the CQ.
3826 * @cq_context: Context associated with the CQ returned to the user via
3827 * the associated completion and event handlers.
3828 * @cq_attr: The attributes the CQ should be created upon.
3830 * Users can examine the cq structure to determine the actual CQ size.
3832 struct ib_cq *__ib_create_cq(struct ib_device *device,
3833 ib_comp_handler comp_handler,
3834 void (*event_handler)(struct ib_event *, void *),
3836 const struct ib_cq_init_attr *cq_attr,
3837 const char *caller);
3838 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3839 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3842 * ib_resize_cq - Modifies the capacity of the CQ.
3843 * @cq: The CQ to resize.
3844 * @cqe: The minimum size of the CQ.
3846 * Users can examine the cq structure to determine the actual CQ size.
3848 int ib_resize_cq(struct ib_cq *cq, int cqe);
3851 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3852 * @cq: The CQ to modify.
3853 * @cq_count: number of CQEs that will trigger an event
3854 * @cq_period: max period of time in usec before triggering an event
3857 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3860 * ib_destroy_cq_user - Destroys the specified CQ.
3861 * @cq: The CQ to destroy.
3862 * @udata: Valid user data or NULL for kernel objects
3864 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3867 * ib_destroy_cq - Destroys the specified kernel CQ.
3868 * @cq: The CQ to destroy.
3870 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3872 static inline void ib_destroy_cq(struct ib_cq *cq)
3874 int ret = ib_destroy_cq_user(cq, NULL);
3876 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3880 * ib_poll_cq - poll a CQ for completion(s)
3881 * @cq:the CQ being polled
3882 * @num_entries:maximum number of completions to return
3883 * @wc:array of at least @num_entries &struct ib_wc where completions
3886 * Poll a CQ for (possibly multiple) completions. If the return value
3887 * is < 0, an error occurred. If the return value is >= 0, it is the
3888 * number of completions returned. If the return value is
3889 * non-negative and < num_entries, then the CQ was emptied.
3891 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3894 return cq->device->ops.poll_cq(cq, num_entries, wc);
3898 * ib_req_notify_cq - Request completion notification on a CQ.
3899 * @cq: The CQ to generate an event for.
3901 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3902 * to request an event on the next solicited event or next work
3903 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3904 * may also be |ed in to request a hint about missed events, as
3908 * < 0 means an error occurred while requesting notification
3909 * == 0 means notification was requested successfully, and if
3910 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3911 * were missed and it is safe to wait for another event. In
3912 * this case is it guaranteed that any work completions added
3913 * to the CQ since the last CQ poll will trigger a completion
3914 * notification event.
3915 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3916 * in. It means that the consumer must poll the CQ again to
3917 * make sure it is empty to avoid missing an event because of a
3918 * race between requesting notification and an entry being
3919 * added to the CQ. This return value means it is possible
3920 * (but not guaranteed) that a work completion has been added
3921 * to the CQ since the last poll without triggering a
3922 * completion notification event.
3924 static inline int ib_req_notify_cq(struct ib_cq *cq,
3925 enum ib_cq_notify_flags flags)
3927 return cq->device->ops.req_notify_cq(cq, flags);
3930 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3931 int comp_vector_hint,
3932 enum ib_poll_context poll_ctx);
3934 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3937 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3938 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3939 * address into the dma address.
3941 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3943 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3947 * ib_dma_mapping_error - check a DMA addr for error
3948 * @dev: The device for which the dma_addr was created
3949 * @dma_addr: The DMA address to check
3951 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3953 if (ib_uses_virt_dma(dev))
3955 return dma_mapping_error(dev->dma_device, dma_addr);
3959 * ib_dma_map_single - Map a kernel virtual address to DMA address
3960 * @dev: The device for which the dma_addr is to be created
3961 * @cpu_addr: The kernel virtual address
3962 * @size: The size of the region in bytes
3963 * @direction: The direction of the DMA
3965 static inline u64 ib_dma_map_single(struct ib_device *dev,
3966 void *cpu_addr, size_t size,
3967 enum dma_data_direction direction)
3969 if (ib_uses_virt_dma(dev))
3970 return (uintptr_t)cpu_addr;
3971 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3975 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3976 * @dev: The device for which the DMA address was created
3977 * @addr: The DMA address
3978 * @size: The size of the region in bytes
3979 * @direction: The direction of the DMA
3981 static inline void ib_dma_unmap_single(struct ib_device *dev,
3982 u64 addr, size_t size,
3983 enum dma_data_direction direction)
3985 if (!ib_uses_virt_dma(dev))
3986 dma_unmap_single(dev->dma_device, addr, size, direction);
3990 * ib_dma_map_page - Map a physical page to DMA address
3991 * @dev: The device for which the dma_addr is to be created
3992 * @page: The page to be mapped
3993 * @offset: The offset within the page
3994 * @size: The size of the region in bytes
3995 * @direction: The direction of the DMA
3997 static inline u64 ib_dma_map_page(struct ib_device *dev,
3999 unsigned long offset,
4001 enum dma_data_direction direction)
4003 if (ib_uses_virt_dma(dev))
4004 return (uintptr_t)(page_address(page) + offset);
4005 return dma_map_page(dev->dma_device, page, offset, size, direction);
4009 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4010 * @dev: The device for which the DMA address was created
4011 * @addr: The DMA address
4012 * @size: The size of the region in bytes
4013 * @direction: The direction of the DMA
4015 static inline void ib_dma_unmap_page(struct ib_device *dev,
4016 u64 addr, size_t size,
4017 enum dma_data_direction direction)
4019 if (!ib_uses_virt_dma(dev))
4020 dma_unmap_page(dev->dma_device, addr, size, direction);
4023 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4024 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4025 struct scatterlist *sg, int nents,
4026 enum dma_data_direction direction,
4027 unsigned long dma_attrs)
4029 if (ib_uses_virt_dma(dev))
4030 return ib_dma_virt_map_sg(dev, sg, nents);
4031 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4035 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4036 struct scatterlist *sg, int nents,
4037 enum dma_data_direction direction,
4038 unsigned long dma_attrs)
4040 if (!ib_uses_virt_dma(dev))
4041 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4046 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4047 * @dev: The device for which the DMA addresses are to be created
4048 * @sg: The array of scatter/gather entries
4049 * @nents: The number of scatter/gather entries
4050 * @direction: The direction of the DMA
4052 static inline int ib_dma_map_sg(struct ib_device *dev,
4053 struct scatterlist *sg, int nents,
4054 enum dma_data_direction direction)
4056 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4060 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4061 * @dev: The device for which the DMA addresses were created
4062 * @sg: The array of scatter/gather entries
4063 * @nents: The number of scatter/gather entries
4064 * @direction: The direction of the DMA
4066 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4067 struct scatterlist *sg, int nents,
4068 enum dma_data_direction direction)
4070 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4074 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4075 * @dev: The device to query
4077 * The returned value represents a size in bytes.
4079 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4081 if (ib_uses_virt_dma(dev))
4083 return dma_get_max_seg_size(dev->dma_device);
4087 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4088 * @dev: The device for which the DMA address was created
4089 * @addr: The DMA address
4090 * @size: The size of the region in bytes
4091 * @dir: The direction of the DMA
4093 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4096 enum dma_data_direction dir)
4098 if (!ib_uses_virt_dma(dev))
4099 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4103 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4104 * @dev: The device for which the DMA address was created
4105 * @addr: The DMA address
4106 * @size: The size of the region in bytes
4107 * @dir: The direction of the DMA
4109 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4112 enum dma_data_direction dir)
4114 if (!ib_uses_virt_dma(dev))
4115 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4118 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4119 * space. This function should be called when 'current' is the owning MM.
4121 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4122 u64 virt_addr, int mr_access_flags);
4124 /* ib_advise_mr - give an advice about an address range in a memory region */
4125 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4126 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4128 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4129 * HCA translation table.
4130 * @mr: The memory region to deregister.
4131 * @udata: Valid user data or NULL for kernel object
4133 * This function can fail, if the memory region has memory windows bound to it.
4135 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4138 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4139 * HCA translation table.
4140 * @mr: The memory region to deregister.
4142 * This function can fail, if the memory region has memory windows bound to it.
4144 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4146 static inline int ib_dereg_mr(struct ib_mr *mr)
4148 return ib_dereg_mr_user(mr, NULL);
4151 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4154 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4155 u32 max_num_data_sg,
4156 u32 max_num_meta_sg);
4159 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4161 * @mr - struct ib_mr pointer to be updated.
4162 * @newkey - new key to be used.
4164 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4166 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4167 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4171 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4172 * for calculating a new rkey for type 2 memory windows.
4173 * @rkey - the rkey to increment.
4175 static inline u32 ib_inc_rkey(u32 rkey)
4177 const u32 mask = 0x000000ff;
4178 return ((rkey + 1) & mask) | (rkey & ~mask);
4182 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4183 * @qp: QP to attach to the multicast group. The QP must be type
4185 * @gid: Multicast group GID.
4186 * @lid: Multicast group LID in host byte order.
4188 * In order to send and receive multicast packets, subnet
4189 * administration must have created the multicast group and configured
4190 * the fabric appropriately. The port associated with the specified
4191 * QP must also be a member of the multicast group.
4193 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4196 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4197 * @qp: QP to detach from the multicast group.
4198 * @gid: Multicast group GID.
4199 * @lid: Multicast group LID in host byte order.
4201 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4203 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4204 struct inode *inode, struct ib_udata *udata);
4205 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4207 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4211 * Local write permission is required if remote write or
4212 * remote atomic permission is also requested.
4214 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4215 !(flags & IB_ACCESS_LOCAL_WRITE))
4218 if (flags & ~IB_ACCESS_SUPPORTED)
4221 if (flags & IB_ACCESS_ON_DEMAND &&
4222 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4227 static inline bool ib_access_writable(int access_flags)
4230 * We have writable memory backing the MR if any of the following
4231 * access flags are set. "Local write" and "remote write" obviously
4232 * require write access. "Remote atomic" can do things like fetch and
4233 * add, which will modify memory, and "MW bind" can change permissions
4234 * by binding a window.
4236 return access_flags &
4237 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4238 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4242 * ib_check_mr_status: lightweight check of MR status.
4243 * This routine may provide status checks on a selected
4244 * ib_mr. first use is for signature status check.
4246 * @mr: A memory region.
4247 * @check_mask: Bitmask of which checks to perform from
4248 * ib_mr_status_check enumeration.
4249 * @mr_status: The container of relevant status checks.
4250 * failed checks will be indicated in the status bitmask
4251 * and the relevant info shall be in the error item.
4253 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4254 struct ib_mr_status *mr_status);
4257 * ib_device_try_get: Hold a registration lock
4258 * device: The device to lock
4260 * A device under an active registration lock cannot become unregistered. It
4261 * is only possible to obtain a registration lock on a device that is fully
4262 * registered, otherwise this function returns false.
4264 * The registration lock is only necessary for actions which require the
4265 * device to still be registered. Uses that only require the device pointer to
4266 * be valid should use get_device(&ibdev->dev) to hold the memory.
4269 static inline bool ib_device_try_get(struct ib_device *dev)
4271 return refcount_inc_not_zero(&dev->refcount);
4274 void ib_device_put(struct ib_device *device);
4275 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4276 enum rdma_driver_id driver_id);
4277 struct ib_device *ib_device_get_by_name(const char *name,
4278 enum rdma_driver_id driver_id);
4279 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4280 u16 pkey, const union ib_gid *gid,
4281 const struct sockaddr *addr);
4282 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4284 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4286 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4287 struct ib_wq_init_attr *init_attr);
4288 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4289 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4292 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4293 unsigned int *sg_offset, unsigned int page_size);
4294 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4295 int data_sg_nents, unsigned int *data_sg_offset,
4296 struct scatterlist *meta_sg, int meta_sg_nents,
4297 unsigned int *meta_sg_offset, unsigned int page_size);
4300 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4301 unsigned int *sg_offset, unsigned int page_size)
4305 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4311 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4312 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4314 void ib_drain_rq(struct ib_qp *qp);
4315 void ib_drain_sq(struct ib_qp *qp);
4316 void ib_drain_qp(struct ib_qp *qp);
4318 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4321 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4323 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4324 return attr->roce.dmac;
4328 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4330 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4331 attr->ib.dlid = (u16)dlid;
4332 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4333 attr->opa.dlid = dlid;
4336 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4338 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4339 return attr->ib.dlid;
4340 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4341 return attr->opa.dlid;
4345 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4350 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4355 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4358 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4359 attr->ib.src_path_bits = src_path_bits;
4360 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4361 attr->opa.src_path_bits = src_path_bits;
4364 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4366 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4367 return attr->ib.src_path_bits;
4368 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4369 return attr->opa.src_path_bits;
4373 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4376 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4377 attr->opa.make_grd = make_grd;
4380 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4382 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4383 return attr->opa.make_grd;
4387 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4389 attr->port_num = port_num;
4392 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4394 return attr->port_num;
4397 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4400 attr->static_rate = static_rate;
4403 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4405 return attr->static_rate;
4408 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4409 enum ib_ah_flags flag)
4411 attr->ah_flags = flag;
4414 static inline enum ib_ah_flags
4415 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4417 return attr->ah_flags;
4420 static inline const struct ib_global_route
4421 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4426 /*To retrieve and modify the grh */
4427 static inline struct ib_global_route
4428 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4433 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4435 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4437 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4440 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4443 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4445 grh->dgid.global.subnet_prefix = prefix;
4448 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4451 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4453 grh->dgid.global.interface_id = if_id;
4456 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4457 union ib_gid *dgid, u32 flow_label,
4458 u8 sgid_index, u8 hop_limit,
4461 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4463 attr->ah_flags = IB_AH_GRH;
4466 grh->flow_label = flow_label;
4467 grh->sgid_index = sgid_index;
4468 grh->hop_limit = hop_limit;
4469 grh->traffic_class = traffic_class;
4470 grh->sgid_attr = NULL;
4473 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4474 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4475 u32 flow_label, u8 hop_limit, u8 traffic_class,
4476 const struct ib_gid_attr *sgid_attr);
4477 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4478 const struct rdma_ah_attr *src);
4479 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4480 const struct rdma_ah_attr *new);
4481 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4484 * rdma_ah_find_type - Return address handle type.
4486 * @dev: Device to be checked
4487 * @port_num: Port number
4489 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4492 if (rdma_protocol_roce(dev, port_num))
4493 return RDMA_AH_ATTR_TYPE_ROCE;
4494 if (rdma_protocol_ib(dev, port_num)) {
4495 if (rdma_cap_opa_ah(dev, port_num))
4496 return RDMA_AH_ATTR_TYPE_OPA;
4497 return RDMA_AH_ATTR_TYPE_IB;
4500 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4504 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4505 * In the current implementation the only way to get
4506 * get the 32bit lid is from other sources for OPA.
4507 * For IB, lids will always be 16bits so cast the
4508 * value accordingly.
4512 static inline u16 ib_lid_cpu16(u32 lid)
4514 WARN_ON_ONCE(lid & 0xFFFF0000);
4519 * ib_lid_be16 - Return lid in 16bit BE encoding.
4523 static inline __be16 ib_lid_be16(u32 lid)
4525 WARN_ON_ONCE(lid & 0xFFFF0000);
4526 return cpu_to_be16((u16)lid);
4530 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4532 * @device: the rdma device
4533 * @comp_vector: index of completion vector
4535 * Returns NULL on failure, otherwise a corresponding cpu map of the
4536 * completion vector (returns all-cpus map if the device driver doesn't
4537 * implement get_vector_affinity).
4539 static inline const struct cpumask *
4540 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4542 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4543 !device->ops.get_vector_affinity)
4546 return device->ops.get_vector_affinity(device, comp_vector);
4551 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4552 * and add their gids, as needed, to the relevant RoCE devices.
4554 * @device: the rdma device
4556 void rdma_roce_rescan_device(struct ib_device *ibdev);
4558 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4560 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4562 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4563 enum rdma_netdev_t type, const char *name,
4564 unsigned char name_assign_type,
4565 void (*setup)(struct net_device *));
4567 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4568 enum rdma_netdev_t type, const char *name,
4569 unsigned char name_assign_type,
4570 void (*setup)(struct net_device *),
4571 struct net_device *netdev);
4574 * rdma_set_device_sysfs_group - Set device attributes group to have
4575 * driver specific sysfs entries at
4576 * for infiniband class.
4578 * @device: device pointer for which attributes to be created
4579 * @group: Pointer to group which should be added when device
4580 * is registered with sysfs.
4581 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4582 * group per device to have sysfs attributes.
4584 * NOTE: New drivers should not make use of this API; instead new device
4585 * parameter should be exposed via netlink command. This API and mechanism
4586 * exist only for existing drivers.
4589 rdma_set_device_sysfs_group(struct ib_device *dev,
4590 const struct attribute_group *group)
4592 dev->groups[1] = group;
4596 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4598 * @device: device pointer for which ib_device pointer to retrieve
4600 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4603 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4605 struct ib_core_device *coredev =
4606 container_of(device, struct ib_core_device, dev);
4608 return coredev->owner;
4612 * ibdev_to_node - return the NUMA node for a given ib_device
4613 * @dev: device to get the NUMA node for.
4615 static inline int ibdev_to_node(struct ib_device *ibdev)
4617 struct device *parent = ibdev->dev.parent;
4620 return NUMA_NO_NODE;
4621 return dev_to_node(parent);
4625 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4626 * ib_device holder structure from device pointer.
4628 * NOTE: New drivers should not make use of this API; This API is only for
4629 * existing drivers who have exposed sysfs entries using
4630 * rdma_set_device_sysfs_group().
4632 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4633 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4635 bool rdma_dev_access_netns(const struct ib_device *device,
4636 const struct net *net);
4638 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4639 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4640 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4643 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4646 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4647 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4650 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4652 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4654 fl_low ^= fl_high >> 14;
4655 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4659 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4660 * local and remote qpn values
4662 * This function folded the multiplication results of two qpns, 24 bit each,
4663 * fields, and converts it to a 20 bit results.
4665 * This function will create symmetric flow_label value based on the local
4666 * and remote qpn values. this will allow both the requester and responder
4667 * to calculate the same flow_label for a given connection.
4669 * This helper function should be used by driver in case the upper layer
4670 * provide a zero flow_label value. This is to improve entropy of RDMA
4671 * traffic in the network.
4673 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4675 u64 v = (u64)lqpn * rqpn;
4680 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4683 const struct ib_port_immutable*
4684 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4685 #endif /* IB_VERBS_H */