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 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/types.h>
16 #include <linux/device.h>
17 #include <linux/dma-mapping.h>
18 #include <linux/kref.h>
19 #include <linux/list.h>
20 #include <linux/rwsem.h>
21 #include <linux/workqueue.h>
22 #include <linux/irq_poll.h>
23 #include <uapi/linux/if_ether.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/netdevice.h>
29 #include <linux/refcount.h>
30 #include <linux/if_link.h>
31 #include <linux/atomic.h>
32 #include <linux/mmu_notifier.h>
33 #include <linux/uaccess.h>
34 #include <linux/cgroup_rdma.h>
35 #include <linux/irqflags.h>
36 #include <linux/preempt.h>
37 #include <linux/dim.h>
38 #include <uapi/rdma/ib_user_verbs.h>
39 #include <rdma/rdma_counter.h>
40 #include <rdma/restrack.h>
41 #include <rdma/signature.h>
42 #include <uapi/rdma/rdma_user_ioctl.h>
43 #include <uapi/rdma/ib_user_ioctl_verbs.h>
45 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
49 struct ib_usrq_object;
53 extern struct workqueue_struct *ib_wq;
54 extern struct workqueue_struct *ib_comp_wq;
55 extern struct workqueue_struct *ib_comp_unbound_wq;
60 void ibdev_printk(const char *level, const struct ib_device *ibdev,
61 const char *format, ...);
63 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
65 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
67 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
69 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
71 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
73 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
75 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
77 #if defined(CONFIG_DYNAMIC_DEBUG) || \
78 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
79 #define ibdev_dbg(__dev, format, args...) \
80 dynamic_ibdev_dbg(__dev, format, ##args)
84 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
87 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
89 static DEFINE_RATELIMIT_STATE(_rs, \
90 DEFAULT_RATELIMIT_INTERVAL, \
91 DEFAULT_RATELIMIT_BURST); \
92 if (__ratelimit(&_rs)) \
93 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
96 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
97 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
98 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
99 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
100 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
101 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
102 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
103 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
104 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
105 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
106 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
107 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
108 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
109 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
111 #if defined(CONFIG_DYNAMIC_DEBUG) || \
112 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
113 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
114 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
116 static DEFINE_RATELIMIT_STATE(_rs, \
117 DEFAULT_RATELIMIT_INTERVAL, \
118 DEFAULT_RATELIMIT_BURST); \
119 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
120 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
121 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
125 __printf(2, 3) __cold
127 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
133 __be64 subnet_prefix;
138 extern union ib_gid zgid;
141 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
142 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
143 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
147 #define ROCE_V2_UDP_DPORT 4791
149 struct net_device __rcu *ndev;
150 struct ib_device *device;
152 enum ib_gid_type gid_type;
158 /* set the local administered indication */
159 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
162 enum rdma_transport_type {
164 RDMA_TRANSPORT_IWARP,
165 RDMA_TRANSPORT_USNIC,
166 RDMA_TRANSPORT_USNIC_UDP,
167 RDMA_TRANSPORT_UNSPECIFIED,
170 enum rdma_protocol_type {
174 RDMA_PROTOCOL_USNIC_UDP
177 __attribute_const__ enum rdma_transport_type
178 rdma_node_get_transport(unsigned int node_type);
180 enum rdma_network_type {
182 RDMA_NETWORK_ROCE_V1,
187 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
189 if (network_type == RDMA_NETWORK_IPV4 ||
190 network_type == RDMA_NETWORK_IPV6)
191 return IB_GID_TYPE_ROCE_UDP_ENCAP;
192 else if (network_type == RDMA_NETWORK_ROCE_V1)
193 return IB_GID_TYPE_ROCE;
195 return IB_GID_TYPE_IB;
198 static inline enum rdma_network_type
199 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
201 if (attr->gid_type == IB_GID_TYPE_IB)
202 return RDMA_NETWORK_IB;
204 if (attr->gid_type == IB_GID_TYPE_ROCE)
205 return RDMA_NETWORK_ROCE_V1;
207 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
208 return RDMA_NETWORK_IPV4;
210 return RDMA_NETWORK_IPV6;
213 enum rdma_link_layer {
214 IB_LINK_LAYER_UNSPECIFIED,
215 IB_LINK_LAYER_INFINIBAND,
216 IB_LINK_LAYER_ETHERNET,
219 enum ib_device_cap_flags {
220 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
221 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
222 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
223 IB_DEVICE_RAW_MULTI = (1 << 3),
224 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
225 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
226 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
227 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
228 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
229 /* Not in use, former INIT_TYPE = (1 << 9),*/
230 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
231 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
232 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
233 IB_DEVICE_SRQ_RESIZE = (1 << 13),
234 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
237 * This device supports a per-device lkey or stag that can be
238 * used without performing a memory registration for the local
239 * memory. Note that ULPs should never check this flag, but
240 * instead of use the local_dma_lkey flag in the ib_pd structure,
241 * which will always contain a usable lkey.
243 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
244 /* Reserved, old SEND_W_INV = (1 << 16),*/
245 IB_DEVICE_MEM_WINDOW = (1 << 17),
247 * Devices should set IB_DEVICE_UD_IP_SUM if they support
248 * insertion of UDP and TCP checksum on outgoing UD IPoIB
249 * messages and can verify the validity of checksum for
250 * incoming messages. Setting this flag implies that the
251 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
253 IB_DEVICE_UD_IP_CSUM = (1 << 18),
254 IB_DEVICE_UD_TSO = (1 << 19),
255 IB_DEVICE_XRC = (1 << 20),
258 * This device supports the IB "base memory management extension",
259 * which includes support for fast registrations (IB_WR_REG_MR,
260 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
261 * also be set by any iWarp device which must support FRs to comply
262 * to the iWarp verbs spec. iWarp devices also support the
263 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
266 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
267 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
268 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
269 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
270 IB_DEVICE_RC_IP_CSUM = (1 << 25),
271 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
272 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
274 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
275 * support execution of WQEs that involve synchronization
276 * of I/O operations with single completion queue managed
279 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
280 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
281 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
282 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
283 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
284 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
285 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
286 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
287 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
288 /* The device supports padding incoming writes to cacheline. */
289 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
290 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
299 enum ib_odp_general_cap_bits {
300 IB_ODP_SUPPORT = 1 << 0,
301 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
304 enum ib_odp_transport_cap_bits {
305 IB_ODP_SUPPORT_SEND = 1 << 0,
306 IB_ODP_SUPPORT_RECV = 1 << 1,
307 IB_ODP_SUPPORT_WRITE = 1 << 2,
308 IB_ODP_SUPPORT_READ = 1 << 3,
309 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
310 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
314 uint64_t general_caps;
316 uint32_t rc_odp_caps;
317 uint32_t uc_odp_caps;
318 uint32_t ud_odp_caps;
319 uint32_t xrc_odp_caps;
320 } per_transport_caps;
324 /* Corresponding bit will be set if qp type from
325 * 'enum ib_qp_type' is supported, e.g.
326 * supported_qpts |= 1 << IB_QPT_UD
329 u32 max_rwq_indirection_tables;
330 u32 max_rwq_indirection_table_size;
333 enum ib_tm_cap_flags {
334 /* Support tag matching with rendezvous offload for RC transport */
335 IB_TM_CAP_RNDV_RC = 1 << 0,
339 /* Max size of RNDV header */
340 u32 max_rndv_hdr_size;
341 /* Max number of entries in tag matching list */
343 /* From enum ib_tm_cap_flags */
345 /* Max number of outstanding list operations */
347 /* Max number of SGE in tag matching entry */
351 struct ib_cq_init_attr {
357 enum ib_cq_attr_mask {
358 IB_CQ_MODERATE = 1 << 0,
362 u16 max_cq_moderation_count;
363 u16 max_cq_moderation_period;
366 struct ib_dm_mr_attr {
372 struct ib_dm_alloc_attr {
378 struct ib_device_attr {
380 __be64 sys_image_guid;
388 u64 device_cap_flags;
399 int max_qp_init_rd_atom;
400 int max_ee_init_rd_atom;
401 enum ib_atomic_cap atomic_cap;
402 enum ib_atomic_cap masked_atomic_cap;
409 int max_mcast_qp_attach;
410 int max_total_mcast_qp_attach;
415 unsigned int max_fast_reg_page_list_len;
416 unsigned int max_pi_fast_reg_page_list_len;
418 u8 local_ca_ack_delay;
421 struct ib_odp_caps odp_caps;
422 uint64_t timestamp_mask;
423 uint64_t hca_core_clock; /* in KHZ */
424 struct ib_rss_caps rss_caps;
426 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
427 struct ib_tm_caps tm_caps;
428 struct ib_cq_caps cq_caps;
430 /* Max entries for sgl for optimized performance per READ */
447 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
450 case IB_MTU_256: return 256;
451 case IB_MTU_512: return 512;
452 case IB_MTU_1024: return 1024;
453 case IB_MTU_2048: return 2048;
454 case IB_MTU_4096: return 4096;
459 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
463 else if (mtu >= 2048)
465 else if (mtu >= 1024)
473 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
481 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
485 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
488 return OPA_MTU_10240;
489 else if (mtu >= 8192)
492 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
501 IB_PORT_ACTIVE_DEFER = 5
504 enum ib_port_phys_state {
505 IB_PORT_PHYS_STATE_SLEEP = 1,
506 IB_PORT_PHYS_STATE_POLLING = 2,
507 IB_PORT_PHYS_STATE_DISABLED = 3,
508 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
509 IB_PORT_PHYS_STATE_LINK_UP = 5,
510 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
511 IB_PORT_PHYS_STATE_PHY_TEST = 7,
522 static inline int ib_width_enum_to_int(enum ib_port_width width)
525 case IB_WIDTH_1X: return 1;
526 case IB_WIDTH_2X: return 2;
527 case IB_WIDTH_4X: return 4;
528 case IB_WIDTH_8X: return 8;
529 case IB_WIDTH_12X: return 12;
546 * struct rdma_hw_stats
547 * @lock - Mutex to protect parallel write access to lifespan and values
548 * of counters, which are 64bits and not guaranteeed to be written
549 * atomicaly on 32bits systems.
550 * @timestamp - Used by the core code to track when the last update was
551 * @lifespan - Used by the core code to determine how old the counters
552 * should be before being updated again. Stored in jiffies, defaults
553 * to 10 milliseconds, drivers can override the default be specifying
554 * their own value during their allocation routine.
555 * @name - Array of pointers to static names used for the counters in
557 * @num_counters - How many hardware counters there are. If name is
558 * shorter than this number, a kernel oops will result. Driver authors
559 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
560 * in their code to prevent this.
561 * @value - Array of u64 counters that are accessed by the sysfs code and
562 * filled in by the drivers get_stats routine
564 struct rdma_hw_stats {
565 struct mutex lock; /* Protect lifespan and values[] */
566 unsigned long timestamp;
567 unsigned long lifespan;
568 const char * const *names;
573 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
575 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
577 * @names - Array of static const char *
578 * @num_counters - How many elements in array
579 * @lifespan - How many milliseconds between updates
581 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
582 const char * const *names, int num_counters,
583 unsigned long lifespan)
585 struct rdma_hw_stats *stats;
587 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
591 stats->names = names;
592 stats->num_counters = num_counters;
593 stats->lifespan = msecs_to_jiffies(lifespan);
599 /* Define bits for the various functionality this port needs to be supported by
602 /* Management 0x00000FFF */
603 #define RDMA_CORE_CAP_IB_MAD 0x00000001
604 #define RDMA_CORE_CAP_IB_SMI 0x00000002
605 #define RDMA_CORE_CAP_IB_CM 0x00000004
606 #define RDMA_CORE_CAP_IW_CM 0x00000008
607 #define RDMA_CORE_CAP_IB_SA 0x00000010
608 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
610 /* Address format 0x000FF000 */
611 #define RDMA_CORE_CAP_AF_IB 0x00001000
612 #define RDMA_CORE_CAP_ETH_AH 0x00002000
613 #define RDMA_CORE_CAP_OPA_AH 0x00004000
614 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
616 /* Protocol 0xFFF00000 */
617 #define RDMA_CORE_CAP_PROT_IB 0x00100000
618 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
619 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
620 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
621 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
622 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
624 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
625 | RDMA_CORE_CAP_PROT_ROCE \
626 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
628 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
629 | RDMA_CORE_CAP_IB_MAD \
630 | RDMA_CORE_CAP_IB_SMI \
631 | RDMA_CORE_CAP_IB_CM \
632 | RDMA_CORE_CAP_IB_SA \
633 | RDMA_CORE_CAP_AF_IB)
634 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
635 | RDMA_CORE_CAP_IB_MAD \
636 | RDMA_CORE_CAP_IB_CM \
637 | RDMA_CORE_CAP_AF_IB \
638 | RDMA_CORE_CAP_ETH_AH)
639 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
640 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
641 | RDMA_CORE_CAP_IB_MAD \
642 | RDMA_CORE_CAP_IB_CM \
643 | RDMA_CORE_CAP_AF_IB \
644 | RDMA_CORE_CAP_ETH_AH)
645 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
646 | RDMA_CORE_CAP_IW_CM)
647 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
648 | RDMA_CORE_CAP_OPA_MAD)
650 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
652 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
654 struct ib_port_attr {
656 enum ib_port_state state;
658 enum ib_mtu active_mtu;
661 unsigned int ip_gids:1;
662 /* This is the value from PortInfo CapabilityMask, defined by IBA */
681 enum ib_device_modify_flags {
682 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
683 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
686 #define IB_DEVICE_NODE_DESC_MAX 64
688 struct ib_device_modify {
690 char node_desc[IB_DEVICE_NODE_DESC_MAX];
693 enum ib_port_modify_flags {
694 IB_PORT_SHUTDOWN = 1,
695 IB_PORT_INIT_TYPE = (1<<2),
696 IB_PORT_RESET_QKEY_CNTR = (1<<3),
697 IB_PORT_OPA_MASK_CHG = (1<<4)
700 struct ib_port_modify {
701 u32 set_port_cap_mask;
702 u32 clr_port_cap_mask;
710 IB_EVENT_QP_ACCESS_ERR,
714 IB_EVENT_PATH_MIG_ERR,
715 IB_EVENT_DEVICE_FATAL,
716 IB_EVENT_PORT_ACTIVE,
719 IB_EVENT_PKEY_CHANGE,
722 IB_EVENT_SRQ_LIMIT_REACHED,
723 IB_EVENT_QP_LAST_WQE_REACHED,
724 IB_EVENT_CLIENT_REREGISTER,
729 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
732 struct ib_device *device;
740 enum ib_event_type event;
743 struct ib_event_handler {
744 struct ib_device *device;
745 void (*handler)(struct ib_event_handler *, struct ib_event *);
746 struct list_head list;
749 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
751 (_ptr)->device = _device; \
752 (_ptr)->handler = _handler; \
753 INIT_LIST_HEAD(&(_ptr)->list); \
756 struct ib_global_route {
757 const struct ib_gid_attr *sgid_attr;
766 __be32 version_tclass_flow;
774 union rdma_network_hdr {
777 /* The IB spec states that if it's IPv4, the header
778 * is located in the last 20 bytes of the header.
781 struct iphdr roce4grh;
785 #define IB_QPN_MASK 0xFFFFFF
788 IB_MULTICAST_QPN = 0xffffff
791 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
792 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
799 IB_RATE_PORT_CURRENT = 0,
800 IB_RATE_2_5_GBPS = 2,
808 IB_RATE_120_GBPS = 10,
809 IB_RATE_14_GBPS = 11,
810 IB_RATE_56_GBPS = 12,
811 IB_RATE_112_GBPS = 13,
812 IB_RATE_168_GBPS = 14,
813 IB_RATE_25_GBPS = 15,
814 IB_RATE_100_GBPS = 16,
815 IB_RATE_200_GBPS = 17,
816 IB_RATE_300_GBPS = 18,
817 IB_RATE_28_GBPS = 19,
818 IB_RATE_50_GBPS = 20,
819 IB_RATE_400_GBPS = 21,
820 IB_RATE_600_GBPS = 22,
824 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
825 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
826 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
827 * @rate: rate to convert.
829 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
832 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
833 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
834 * @rate: rate to convert.
836 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
840 * enum ib_mr_type - memory region type
841 * @IB_MR_TYPE_MEM_REG: memory region that is used for
842 * normal registration
843 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
844 * register any arbitrary sg lists (without
845 * the normal mr constraints - see
847 * @IB_MR_TYPE_DM: memory region that is used for device
848 * memory registration
849 * @IB_MR_TYPE_USER: memory region that is used for the user-space
851 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
852 * without address translations (VA=PA)
853 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
854 * data integrity operations
862 IB_MR_TYPE_INTEGRITY,
865 enum ib_mr_status_check {
866 IB_MR_CHECK_SIG_STATUS = 1,
870 * struct ib_mr_status - Memory region status container
872 * @fail_status: Bitmask of MR checks status. For each
873 * failed check a corresponding status bit is set.
874 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
877 struct ib_mr_status {
879 struct ib_sig_err sig_err;
883 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
885 * @mult: multiple to convert.
887 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
889 struct rdma_ah_init_attr {
890 struct rdma_ah_attr *ah_attr;
892 struct net_device *xmit_slave;
895 enum rdma_ah_attr_type {
896 RDMA_AH_ATTR_TYPE_UNDEFINED,
897 RDMA_AH_ATTR_TYPE_IB,
898 RDMA_AH_ATTR_TYPE_ROCE,
899 RDMA_AH_ATTR_TYPE_OPA,
907 struct roce_ah_attr {
917 struct rdma_ah_attr {
918 struct ib_global_route grh;
923 enum rdma_ah_attr_type type;
925 struct ib_ah_attr ib;
926 struct roce_ah_attr roce;
927 struct opa_ah_attr opa;
935 IB_WC_LOC_EEC_OP_ERR,
940 IB_WC_LOC_ACCESS_ERR,
941 IB_WC_REM_INV_REQ_ERR,
942 IB_WC_REM_ACCESS_ERR,
945 IB_WC_RNR_RETRY_EXC_ERR,
946 IB_WC_LOC_RDD_VIOL_ERR,
947 IB_WC_REM_INV_RD_REQ_ERR,
950 IB_WC_INV_EEC_STATE_ERR,
952 IB_WC_RESP_TIMEOUT_ERR,
956 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
959 IB_WC_SEND = IB_UVERBS_WC_SEND,
960 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
961 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
962 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
963 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
964 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
965 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
966 IB_WC_LSO = IB_UVERBS_WC_TSO,
968 IB_WC_MASKED_COMP_SWAP,
969 IB_WC_MASKED_FETCH_ADD,
971 * Set value of IB_WC_RECV so consumers can test if a completion is a
972 * receive by testing (opcode & IB_WC_RECV).
975 IB_WC_RECV_RDMA_WITH_IMM
980 IB_WC_WITH_IMM = (1<<1),
981 IB_WC_WITH_INVALIDATE = (1<<2),
982 IB_WC_IP_CSUM_OK = (1<<3),
983 IB_WC_WITH_SMAC = (1<<4),
984 IB_WC_WITH_VLAN = (1<<5),
985 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
991 struct ib_cqe *wr_cqe;
993 enum ib_wc_status status;
994 enum ib_wc_opcode opcode;
1000 u32 invalidate_rkey;
1008 u8 port_num; /* valid only for DR SMPs on switches */
1011 u8 network_hdr_type;
1014 enum ib_cq_notify_flags {
1015 IB_CQ_SOLICITED = 1 << 0,
1016 IB_CQ_NEXT_COMP = 1 << 1,
1017 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1018 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1022 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1023 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1024 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1027 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1029 return srq_type == IB_SRQT_XRC ||
1030 srq_type == IB_SRQT_TM;
1033 enum ib_srq_attr_mask {
1034 IB_SRQ_MAX_WR = 1 << 0,
1035 IB_SRQ_LIMIT = 1 << 1,
1038 struct ib_srq_attr {
1044 struct ib_srq_init_attr {
1045 void (*event_handler)(struct ib_event *, void *);
1047 struct ib_srq_attr attr;
1048 enum ib_srq_type srq_type;
1054 struct ib_xrcd *xrcd;
1069 u32 max_inline_data;
1072 * Maximum number of rdma_rw_ctx structures in flight at a time.
1073 * ib_create_qp() will calculate the right amount of neededed WRs
1074 * and MRs based on this.
1086 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1087 * here (and in that order) since the MAD layer uses them as
1088 * indices into a 2-entry table.
1093 IB_QPT_RC = IB_UVERBS_QPT_RC,
1094 IB_QPT_UC = IB_UVERBS_QPT_UC,
1095 IB_QPT_UD = IB_UVERBS_QPT_UD,
1097 IB_QPT_RAW_ETHERTYPE,
1098 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1099 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1100 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1102 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1103 /* Reserve a range for qp types internal to the low level driver.
1104 * These qp types will not be visible at the IB core layer, so the
1105 * IB_QPT_MAX usages should not be affected in the core layer
1107 IB_QPT_RESERVED1 = 0x1000,
1119 enum ib_qp_create_flags {
1120 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1121 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1122 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1123 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1124 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1125 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1126 IB_QP_CREATE_NETIF_QP = 1 << 5,
1127 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1128 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1129 IB_QP_CREATE_SCATTER_FCS =
1130 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1131 IB_QP_CREATE_CVLAN_STRIPPING =
1132 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1133 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1134 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1135 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1136 /* reserve bits 26-31 for low level drivers' internal use */
1137 IB_QP_CREATE_RESERVED_START = 1 << 26,
1138 IB_QP_CREATE_RESERVED_END = 1 << 31,
1142 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1143 * callback to destroy the passed in QP.
1146 struct ib_qp_init_attr {
1147 /* Consumer's event_handler callback must not block */
1148 void (*event_handler)(struct ib_event *, void *);
1151 struct ib_cq *send_cq;
1152 struct ib_cq *recv_cq;
1154 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1155 struct ib_qp_cap cap;
1156 enum ib_sig_type sq_sig_type;
1157 enum ib_qp_type qp_type;
1161 * Only needed for special QP types, or when using the RW API.
1164 struct ib_rwq_ind_table *rwq_ind_tbl;
1168 struct ib_qp_open_attr {
1169 void (*event_handler)(struct ib_event *, void *);
1172 enum ib_qp_type qp_type;
1175 enum ib_rnr_timeout {
1176 IB_RNR_TIMER_655_36 = 0,
1177 IB_RNR_TIMER_000_01 = 1,
1178 IB_RNR_TIMER_000_02 = 2,
1179 IB_RNR_TIMER_000_03 = 3,
1180 IB_RNR_TIMER_000_04 = 4,
1181 IB_RNR_TIMER_000_06 = 5,
1182 IB_RNR_TIMER_000_08 = 6,
1183 IB_RNR_TIMER_000_12 = 7,
1184 IB_RNR_TIMER_000_16 = 8,
1185 IB_RNR_TIMER_000_24 = 9,
1186 IB_RNR_TIMER_000_32 = 10,
1187 IB_RNR_TIMER_000_48 = 11,
1188 IB_RNR_TIMER_000_64 = 12,
1189 IB_RNR_TIMER_000_96 = 13,
1190 IB_RNR_TIMER_001_28 = 14,
1191 IB_RNR_TIMER_001_92 = 15,
1192 IB_RNR_TIMER_002_56 = 16,
1193 IB_RNR_TIMER_003_84 = 17,
1194 IB_RNR_TIMER_005_12 = 18,
1195 IB_RNR_TIMER_007_68 = 19,
1196 IB_RNR_TIMER_010_24 = 20,
1197 IB_RNR_TIMER_015_36 = 21,
1198 IB_RNR_TIMER_020_48 = 22,
1199 IB_RNR_TIMER_030_72 = 23,
1200 IB_RNR_TIMER_040_96 = 24,
1201 IB_RNR_TIMER_061_44 = 25,
1202 IB_RNR_TIMER_081_92 = 26,
1203 IB_RNR_TIMER_122_88 = 27,
1204 IB_RNR_TIMER_163_84 = 28,
1205 IB_RNR_TIMER_245_76 = 29,
1206 IB_RNR_TIMER_327_68 = 30,
1207 IB_RNR_TIMER_491_52 = 31
1210 enum ib_qp_attr_mask {
1212 IB_QP_CUR_STATE = (1<<1),
1213 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1214 IB_QP_ACCESS_FLAGS = (1<<3),
1215 IB_QP_PKEY_INDEX = (1<<4),
1216 IB_QP_PORT = (1<<5),
1217 IB_QP_QKEY = (1<<6),
1219 IB_QP_PATH_MTU = (1<<8),
1220 IB_QP_TIMEOUT = (1<<9),
1221 IB_QP_RETRY_CNT = (1<<10),
1222 IB_QP_RNR_RETRY = (1<<11),
1223 IB_QP_RQ_PSN = (1<<12),
1224 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1225 IB_QP_ALT_PATH = (1<<14),
1226 IB_QP_MIN_RNR_TIMER = (1<<15),
1227 IB_QP_SQ_PSN = (1<<16),
1228 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1229 IB_QP_PATH_MIG_STATE = (1<<18),
1230 IB_QP_CAP = (1<<19),
1231 IB_QP_DEST_QPN = (1<<20),
1232 IB_QP_RESERVED1 = (1<<21),
1233 IB_QP_RESERVED2 = (1<<22),
1234 IB_QP_RESERVED3 = (1<<23),
1235 IB_QP_RESERVED4 = (1<<24),
1236 IB_QP_RATE_LIMIT = (1<<25),
1238 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1263 enum ib_qp_state qp_state;
1264 enum ib_qp_state cur_qp_state;
1265 enum ib_mtu path_mtu;
1266 enum ib_mig_state path_mig_state;
1271 int qp_access_flags;
1272 struct ib_qp_cap cap;
1273 struct rdma_ah_attr ah_attr;
1274 struct rdma_ah_attr alt_ah_attr;
1277 u8 en_sqd_async_notify;
1280 u8 max_dest_rd_atomic;
1289 struct net_device *xmit_slave;
1293 /* These are shared with userspace */
1294 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1295 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1296 IB_WR_SEND = IB_UVERBS_WR_SEND,
1297 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1298 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1299 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1300 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1301 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1302 IB_WR_LSO = IB_UVERBS_WR_TSO,
1303 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1304 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1305 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1306 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1307 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1308 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1309 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1311 /* These are kernel only and can not be issued by userspace */
1312 IB_WR_REG_MR = 0x20,
1313 IB_WR_REG_MR_INTEGRITY,
1315 /* reserve values for low level drivers' internal use.
1316 * These values will not be used at all in the ib core layer.
1318 IB_WR_RESERVED1 = 0xf0,
1330 enum ib_send_flags {
1332 IB_SEND_SIGNALED = (1<<1),
1333 IB_SEND_SOLICITED = (1<<2),
1334 IB_SEND_INLINE = (1<<3),
1335 IB_SEND_IP_CSUM = (1<<4),
1337 /* reserve bits 26-31 for low level drivers' internal use */
1338 IB_SEND_RESERVED_START = (1 << 26),
1339 IB_SEND_RESERVED_END = (1 << 31),
1349 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1353 struct ib_send_wr *next;
1356 struct ib_cqe *wr_cqe;
1358 struct ib_sge *sg_list;
1360 enum ib_wr_opcode opcode;
1364 u32 invalidate_rkey;
1369 struct ib_send_wr wr;
1374 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1376 return container_of(wr, struct ib_rdma_wr, wr);
1379 struct ib_atomic_wr {
1380 struct ib_send_wr wr;
1384 u64 compare_add_mask;
1389 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1391 return container_of(wr, struct ib_atomic_wr, wr);
1395 struct ib_send_wr wr;
1402 u16 pkey_index; /* valid for GSI only */
1403 u8 port_num; /* valid for DR SMPs on switch only */
1406 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1408 return container_of(wr, struct ib_ud_wr, wr);
1412 struct ib_send_wr wr;
1418 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1420 return container_of(wr, struct ib_reg_wr, wr);
1424 struct ib_recv_wr *next;
1427 struct ib_cqe *wr_cqe;
1429 struct ib_sge *sg_list;
1433 enum ib_access_flags {
1434 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1435 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1436 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1437 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1438 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1439 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1440 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1441 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1442 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1444 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1445 IB_ACCESS_SUPPORTED =
1446 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1450 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1451 * are hidden here instead of a uapi header!
1453 enum ib_mr_rereg_flags {
1454 IB_MR_REREG_TRANS = 1,
1455 IB_MR_REREG_PD = (1<<1),
1456 IB_MR_REREG_ACCESS = (1<<2),
1457 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1462 enum rdma_remove_reason {
1464 * Userspace requested uobject deletion or initial try
1465 * to remove uobject via cleanup. Call could fail
1467 RDMA_REMOVE_DESTROY,
1468 /* Context deletion. This call should delete the actual object itself */
1470 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1471 RDMA_REMOVE_DRIVER_REMOVE,
1472 /* uobj is being cleaned-up before being committed */
1474 /* The driver failed to destroy the uobject and is being disconnected */
1475 RDMA_REMOVE_DRIVER_FAILURE,
1478 struct ib_rdmacg_object {
1479 #ifdef CONFIG_CGROUP_RDMA
1480 struct rdma_cgroup *cg; /* owner rdma cgroup */
1484 struct ib_ucontext {
1485 struct ib_device *device;
1486 struct ib_uverbs_file *ufile;
1488 struct ib_rdmacg_object cg_obj;
1490 * Implementation details of the RDMA core, don't use in drivers:
1492 struct rdma_restrack_entry res;
1493 struct xarray mmap_xa;
1497 u64 user_handle; /* handle given to us by userspace */
1498 /* ufile & ucontext owning this object */
1499 struct ib_uverbs_file *ufile;
1500 /* FIXME, save memory: ufile->context == context */
1501 struct ib_ucontext *context; /* associated user context */
1502 void *object; /* containing object */
1503 struct list_head list; /* link to context's list */
1504 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1505 int id; /* index into kernel idr */
1507 atomic_t usecnt; /* protects exclusive access */
1508 struct rcu_head rcu; /* kfree_rcu() overhead */
1510 const struct uverbs_api_object *uapi_object;
1514 const void __user *inbuf;
1515 void __user *outbuf;
1523 struct ib_device *device;
1524 struct ib_uobject *uobject;
1525 atomic_t usecnt; /* count all resources */
1527 u32 unsafe_global_rkey;
1530 * Implementation details of the RDMA core, don't use in drivers:
1532 struct ib_mr *__internal_mr;
1533 struct rdma_restrack_entry res;
1537 struct ib_device *device;
1538 atomic_t usecnt; /* count all exposed resources */
1539 struct inode *inode;
1540 struct rw_semaphore tgt_qps_rwsem;
1541 struct xarray tgt_qps;
1545 struct ib_device *device;
1547 struct ib_uobject *uobject;
1548 const struct ib_gid_attr *sgid_attr;
1549 enum rdma_ah_attr_type type;
1552 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1554 enum ib_poll_context {
1555 IB_POLL_SOFTIRQ, /* poll from softirq context */
1556 IB_POLL_WORKQUEUE, /* poll from workqueue */
1557 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1558 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1560 IB_POLL_DIRECT, /* caller context, no hw completions */
1564 struct ib_device *device;
1565 struct ib_ucq_object *uobject;
1566 ib_comp_handler comp_handler;
1567 void (*event_handler)(struct ib_event *, void *);
1570 unsigned int cqe_used;
1571 atomic_t usecnt; /* count number of work queues */
1572 enum ib_poll_context poll_ctx;
1574 struct list_head pool_entry;
1576 struct irq_poll iop;
1577 struct work_struct work;
1579 struct workqueue_struct *comp_wq;
1582 /* updated only by trace points */
1586 unsigned int comp_vector;
1589 * Implementation details of the RDMA core, don't use in drivers:
1591 struct rdma_restrack_entry res;
1595 struct ib_device *device;
1597 struct ib_usrq_object *uobject;
1598 void (*event_handler)(struct ib_event *, void *);
1600 enum ib_srq_type srq_type;
1607 struct ib_xrcd *xrcd;
1614 enum ib_raw_packet_caps {
1615 /* Strip cvlan from incoming packet and report it in the matching work
1616 * completion is supported.
1618 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1619 /* Scatter FCS field of an incoming packet to host memory is supported.
1621 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1622 /* Checksum offloads are supported (for both send and receive). */
1623 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1624 /* When a packet is received for an RQ with no receive WQEs, the
1625 * packet processing is delayed.
1627 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1631 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1641 struct ib_device *device;
1642 struct ib_uwq_object *uobject;
1644 void (*event_handler)(struct ib_event *, void *);
1648 enum ib_wq_state state;
1649 enum ib_wq_type wq_type;
1654 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1655 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1656 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1657 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1658 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1661 struct ib_wq_init_attr {
1663 enum ib_wq_type wq_type;
1667 void (*event_handler)(struct ib_event *, void *);
1668 u32 create_flags; /* Use enum ib_wq_flags */
1671 enum ib_wq_attr_mask {
1672 IB_WQ_STATE = 1 << 0,
1673 IB_WQ_CUR_STATE = 1 << 1,
1674 IB_WQ_FLAGS = 1 << 2,
1678 enum ib_wq_state wq_state;
1679 enum ib_wq_state curr_wq_state;
1680 u32 flags; /* Use enum ib_wq_flags */
1681 u32 flags_mask; /* Use enum ib_wq_flags */
1684 struct ib_rwq_ind_table {
1685 struct ib_device *device;
1686 struct ib_uobject *uobject;
1689 u32 log_ind_tbl_size;
1690 struct ib_wq **ind_tbl;
1693 struct ib_rwq_ind_table_init_attr {
1694 u32 log_ind_tbl_size;
1695 /* Each entry is a pointer to Receive Work Queue */
1696 struct ib_wq **ind_tbl;
1699 enum port_pkey_state {
1700 IB_PORT_PKEY_NOT_VALID = 0,
1701 IB_PORT_PKEY_VALID = 1,
1702 IB_PORT_PKEY_LISTED = 2,
1705 struct ib_qp_security;
1707 struct ib_port_pkey {
1708 enum port_pkey_state state;
1711 struct list_head qp_list;
1712 struct list_head to_error_list;
1713 struct ib_qp_security *sec;
1716 struct ib_ports_pkeys {
1717 struct ib_port_pkey main;
1718 struct ib_port_pkey alt;
1721 struct ib_qp_security {
1723 struct ib_device *dev;
1724 /* Hold this mutex when changing port and pkey settings. */
1726 struct ib_ports_pkeys *ports_pkeys;
1727 /* A list of all open shared QP handles. Required to enforce security
1728 * properly for all users of a shared QP.
1730 struct list_head shared_qp_list;
1733 atomic_t error_list_count;
1734 struct completion error_complete;
1735 int error_comps_pending;
1739 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1740 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1743 struct ib_device *device;
1745 struct ib_cq *send_cq;
1746 struct ib_cq *recv_cq;
1749 struct list_head rdma_mrs;
1750 struct list_head sig_mrs;
1752 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1753 struct list_head xrcd_list;
1755 /* count times opened, mcast attaches, flow attaches */
1757 struct list_head open_list;
1758 struct ib_qp *real_qp;
1759 struct ib_uqp_object *uobject;
1760 void (*event_handler)(struct ib_event *, void *);
1762 /* sgid_attrs associated with the AV's */
1763 const struct ib_gid_attr *av_sgid_attr;
1764 const struct ib_gid_attr *alt_path_sgid_attr;
1768 enum ib_qp_type qp_type;
1769 struct ib_rwq_ind_table *rwq_ind_tbl;
1770 struct ib_qp_security *qp_sec;
1775 * Implementation details of the RDMA core, don't use in drivers:
1777 struct rdma_restrack_entry res;
1779 /* The counter the qp is bind to */
1780 struct rdma_counter *counter;
1784 struct ib_device *device;
1787 struct ib_uobject *uobject;
1792 struct ib_device *device;
1798 unsigned int page_size;
1799 enum ib_mr_type type;
1802 struct ib_uobject *uobject; /* user */
1803 struct list_head qp_entry; /* FR */
1807 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1809 * Implementation details of the RDMA core, don't use in drivers:
1811 struct rdma_restrack_entry res;
1815 struct ib_device *device;
1817 struct ib_uobject *uobject;
1819 enum ib_mw_type type;
1822 /* Supported steering options */
1823 enum ib_flow_attr_type {
1824 /* steering according to rule specifications */
1825 IB_FLOW_ATTR_NORMAL = 0x0,
1826 /* default unicast and multicast rule -
1827 * receive all Eth traffic which isn't steered to any QP
1829 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1830 /* default multicast rule -
1831 * receive all Eth multicast traffic which isn't steered to any QP
1833 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1834 /* sniffer rule - receive all port traffic */
1835 IB_FLOW_ATTR_SNIFFER = 0x3
1838 /* Supported steering header types */
1839 enum ib_flow_spec_type {
1841 IB_FLOW_SPEC_ETH = 0x20,
1842 IB_FLOW_SPEC_IB = 0x22,
1844 IB_FLOW_SPEC_IPV4 = 0x30,
1845 IB_FLOW_SPEC_IPV6 = 0x31,
1846 IB_FLOW_SPEC_ESP = 0x34,
1848 IB_FLOW_SPEC_TCP = 0x40,
1849 IB_FLOW_SPEC_UDP = 0x41,
1850 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1851 IB_FLOW_SPEC_GRE = 0x51,
1852 IB_FLOW_SPEC_MPLS = 0x60,
1853 IB_FLOW_SPEC_INNER = 0x100,
1855 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1856 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1857 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1858 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1860 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1861 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1863 enum ib_flow_flags {
1864 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1865 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1866 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1869 struct ib_flow_eth_filter {
1878 struct ib_flow_spec_eth {
1881 struct ib_flow_eth_filter val;
1882 struct ib_flow_eth_filter mask;
1885 struct ib_flow_ib_filter {
1892 struct ib_flow_spec_ib {
1895 struct ib_flow_ib_filter val;
1896 struct ib_flow_ib_filter mask;
1899 /* IPv4 header flags */
1900 enum ib_ipv4_flags {
1901 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1902 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1903 last have this flag set */
1906 struct ib_flow_ipv4_filter {
1917 struct ib_flow_spec_ipv4 {
1920 struct ib_flow_ipv4_filter val;
1921 struct ib_flow_ipv4_filter mask;
1924 struct ib_flow_ipv6_filter {
1935 struct ib_flow_spec_ipv6 {
1938 struct ib_flow_ipv6_filter val;
1939 struct ib_flow_ipv6_filter mask;
1942 struct ib_flow_tcp_udp_filter {
1949 struct ib_flow_spec_tcp_udp {
1952 struct ib_flow_tcp_udp_filter val;
1953 struct ib_flow_tcp_udp_filter mask;
1956 struct ib_flow_tunnel_filter {
1961 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1962 * the tunnel_id from val has the vni value
1964 struct ib_flow_spec_tunnel {
1967 struct ib_flow_tunnel_filter val;
1968 struct ib_flow_tunnel_filter mask;
1971 struct ib_flow_esp_filter {
1978 struct ib_flow_spec_esp {
1981 struct ib_flow_esp_filter val;
1982 struct ib_flow_esp_filter mask;
1985 struct ib_flow_gre_filter {
1986 __be16 c_ks_res0_ver;
1993 struct ib_flow_spec_gre {
1996 struct ib_flow_gre_filter val;
1997 struct ib_flow_gre_filter mask;
2000 struct ib_flow_mpls_filter {
2006 struct ib_flow_spec_mpls {
2009 struct ib_flow_mpls_filter val;
2010 struct ib_flow_mpls_filter mask;
2013 struct ib_flow_spec_action_tag {
2014 enum ib_flow_spec_type type;
2019 struct ib_flow_spec_action_drop {
2020 enum ib_flow_spec_type type;
2024 struct ib_flow_spec_action_handle {
2025 enum ib_flow_spec_type type;
2027 struct ib_flow_action *act;
2030 enum ib_counters_description {
2035 struct ib_flow_spec_action_count {
2036 enum ib_flow_spec_type type;
2038 struct ib_counters *counters;
2041 union ib_flow_spec {
2046 struct ib_flow_spec_eth eth;
2047 struct ib_flow_spec_ib ib;
2048 struct ib_flow_spec_ipv4 ipv4;
2049 struct ib_flow_spec_tcp_udp tcp_udp;
2050 struct ib_flow_spec_ipv6 ipv6;
2051 struct ib_flow_spec_tunnel tunnel;
2052 struct ib_flow_spec_esp esp;
2053 struct ib_flow_spec_gre gre;
2054 struct ib_flow_spec_mpls mpls;
2055 struct ib_flow_spec_action_tag flow_tag;
2056 struct ib_flow_spec_action_drop drop;
2057 struct ib_flow_spec_action_handle action;
2058 struct ib_flow_spec_action_count flow_count;
2061 struct ib_flow_attr {
2062 enum ib_flow_attr_type type;
2068 union ib_flow_spec flows[];
2073 struct ib_device *device;
2074 struct ib_uobject *uobject;
2077 enum ib_flow_action_type {
2078 IB_FLOW_ACTION_UNSPECIFIED,
2079 IB_FLOW_ACTION_ESP = 1,
2082 struct ib_flow_action_attrs_esp_keymats {
2083 enum ib_uverbs_flow_action_esp_keymat protocol;
2085 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2089 struct ib_flow_action_attrs_esp_replays {
2090 enum ib_uverbs_flow_action_esp_replay protocol;
2092 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2096 enum ib_flow_action_attrs_esp_flags {
2097 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2098 * This is done in order to share the same flags between user-space and
2099 * kernel and spare an unnecessary translation.
2103 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2104 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2107 struct ib_flow_spec_list {
2108 struct ib_flow_spec_list *next;
2109 union ib_flow_spec spec;
2112 struct ib_flow_action_attrs_esp {
2113 struct ib_flow_action_attrs_esp_keymats *keymat;
2114 struct ib_flow_action_attrs_esp_replays *replay;
2115 struct ib_flow_spec_list *encap;
2116 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2117 * Value of 0 is a valid value.
2123 /* Use enum ib_flow_action_attrs_esp_flags */
2125 u64 hard_limit_pkts;
2128 struct ib_flow_action {
2129 struct ib_device *device;
2130 struct ib_uobject *uobject;
2131 enum ib_flow_action_type type;
2138 enum ib_process_mad_flags {
2139 IB_MAD_IGNORE_MKEY = 1,
2140 IB_MAD_IGNORE_BKEY = 2,
2141 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2144 enum ib_mad_result {
2145 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2146 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2147 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2148 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2151 struct ib_port_cache {
2153 struct ib_pkey_cache *pkey;
2154 struct ib_gid_table *gid;
2156 enum ib_port_state port_state;
2159 struct ib_port_immutable {
2166 struct ib_port_data {
2167 struct ib_device *ib_dev;
2169 struct ib_port_immutable immutable;
2171 spinlock_t pkey_list_lock;
2172 struct list_head pkey_list;
2174 struct ib_port_cache cache;
2176 spinlock_t netdev_lock;
2177 struct net_device __rcu *netdev;
2178 struct hlist_node ndev_hash_link;
2179 struct rdma_port_counter port_counter;
2180 struct rdma_hw_stats *hw_stats;
2183 /* rdma netdev type - specifies protocol type */
2184 enum rdma_netdev_t {
2185 RDMA_NETDEV_OPA_VNIC,
2190 * struct rdma_netdev - rdma netdev
2191 * For cases where netstack interfacing is required.
2193 struct rdma_netdev {
2195 struct ib_device *hca;
2200 * cleanup function must be specified.
2201 * FIXME: This is only used for OPA_VNIC and that usage should be
2204 void (*free_rdma_netdev)(struct net_device *netdev);
2206 /* control functions */
2207 void (*set_id)(struct net_device *netdev, int id);
2209 int (*send)(struct net_device *dev, struct sk_buff *skb,
2210 struct ib_ah *address, u32 dqpn);
2212 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2213 union ib_gid *gid, u16 mlid,
2214 int set_qkey, u32 qkey);
2215 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2216 union ib_gid *gid, u16 mlid);
2219 struct rdma_netdev_alloc_params {
2225 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2226 struct net_device *netdev, void *param);
2229 struct ib_odp_counters {
2231 atomic64_t invalidations;
2232 atomic64_t prefetch;
2235 struct ib_counters {
2236 struct ib_device *device;
2237 struct ib_uobject *uobject;
2238 /* num of objects attached */
2242 struct ib_counters_read_attr {
2245 u32 flags; /* use enum ib_read_counters_flags */
2248 struct uverbs_attr_bundle;
2250 struct iw_cm_conn_param;
2252 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2253 .size_##ib_struct = \
2254 (sizeof(struct drv_struct) + \
2255 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2256 BUILD_BUG_ON_ZERO( \
2257 !__same_type(((struct drv_struct *)NULL)->member, \
2260 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2261 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2263 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2264 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2266 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2268 struct rdma_user_mmap_entry {
2270 struct ib_ucontext *ucontext;
2271 unsigned long start_pgoff;
2273 bool driver_removed;
2276 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2278 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2280 return (u64)entry->start_pgoff << PAGE_SHIFT;
2284 * struct ib_device_ops - InfiniBand device operations
2285 * This structure defines all the InfiniBand device operations, providers will
2286 * need to define the supported operations, otherwise they will be set to null.
2288 struct ib_device_ops {
2289 struct module *owner;
2290 enum rdma_driver_id driver_id;
2292 unsigned int uverbs_no_driver_id_binding:1;
2294 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2295 const struct ib_send_wr **bad_send_wr);
2296 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2297 const struct ib_recv_wr **bad_recv_wr);
2298 void (*drain_rq)(struct ib_qp *qp);
2299 void (*drain_sq)(struct ib_qp *qp);
2300 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2301 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2302 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2303 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2304 int (*post_srq_recv)(struct ib_srq *srq,
2305 const struct ib_recv_wr *recv_wr,
2306 const struct ib_recv_wr **bad_recv_wr);
2307 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2308 u8 port_num, const struct ib_wc *in_wc,
2309 const struct ib_grh *in_grh,
2310 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2311 size_t *out_mad_size, u16 *out_mad_pkey_index);
2312 int (*query_device)(struct ib_device *device,
2313 struct ib_device_attr *device_attr,
2314 struct ib_udata *udata);
2315 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2316 struct ib_device_modify *device_modify);
2317 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2318 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2320 int (*query_port)(struct ib_device *device, u8 port_num,
2321 struct ib_port_attr *port_attr);
2322 int (*modify_port)(struct ib_device *device, u8 port_num,
2323 int port_modify_mask,
2324 struct ib_port_modify *port_modify);
2326 * The following mandatory functions are used only at device
2327 * registration. Keep functions such as these at the end of this
2328 * structure to avoid cache line misses when accessing struct ib_device
2331 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2332 struct ib_port_immutable *immutable);
2333 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2336 * When calling get_netdev, the HW vendor's driver should return the
2337 * net device of device @device at port @port_num or NULL if such
2338 * a net device doesn't exist. The vendor driver should call dev_hold
2339 * on this net device. The HW vendor's device driver must guarantee
2340 * that this function returns NULL before the net device has finished
2341 * NETDEV_UNREGISTER state.
2343 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2345 * rdma netdev operation
2347 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2348 * must return -EOPNOTSUPP if it doesn't support the specified type.
2350 struct net_device *(*alloc_rdma_netdev)(
2351 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2352 const char *name, unsigned char name_assign_type,
2353 void (*setup)(struct net_device *));
2355 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2356 enum rdma_netdev_t type,
2357 struct rdma_netdev_alloc_params *params);
2359 * query_gid should be return GID value for @device, when @port_num
2360 * link layer is either IB or iWarp. It is no-op if @port_num port
2361 * is RoCE link layer.
2363 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2366 * When calling add_gid, the HW vendor's driver should add the gid
2367 * of device of port at gid index available at @attr. Meta-info of
2368 * that gid (for example, the network device related to this gid) is
2369 * available at @attr. @context allows the HW vendor driver to store
2370 * extra information together with a GID entry. The HW vendor driver may
2371 * allocate memory to contain this information and store it in @context
2372 * when a new GID entry is written to. Params are consistent until the
2373 * next call of add_gid or delete_gid. The function should return 0 on
2374 * success or error otherwise. The function could be called
2375 * concurrently for different ports. This function is only called when
2376 * roce_gid_table is used.
2378 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2380 * When calling del_gid, the HW vendor's driver should delete the
2381 * gid of device @device at gid index gid_index of port port_num
2382 * available in @attr.
2383 * Upon the deletion of a GID entry, the HW vendor must free any
2384 * allocated memory. The caller will clear @context afterwards.
2385 * This function is only called when roce_gid_table is used.
2387 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2388 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2390 int (*alloc_ucontext)(struct ib_ucontext *context,
2391 struct ib_udata *udata);
2392 void (*dealloc_ucontext)(struct ib_ucontext *context);
2393 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2395 * This will be called once refcount of an entry in mmap_xa reaches
2396 * zero. The type of the memory that was mapped may differ between
2397 * entries and is opaque to the rdma_user_mmap interface.
2398 * Therefore needs to be implemented by the driver in mmap_free.
2400 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2401 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2402 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2403 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2404 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2405 struct ib_udata *udata);
2406 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2407 struct ib_udata *udata);
2408 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2409 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2410 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2411 int (*create_srq)(struct ib_srq *srq,
2412 struct ib_srq_init_attr *srq_init_attr,
2413 struct ib_udata *udata);
2414 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2415 enum ib_srq_attr_mask srq_attr_mask,
2416 struct ib_udata *udata);
2417 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2418 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2419 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2420 struct ib_qp_init_attr *qp_init_attr,
2421 struct ib_udata *udata);
2422 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2423 int qp_attr_mask, struct ib_udata *udata);
2424 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2425 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2426 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2427 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2428 struct ib_udata *udata);
2429 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2430 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2431 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2432 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2433 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2434 u64 virt_addr, int mr_access_flags,
2435 struct ib_udata *udata);
2436 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2437 u64 virt_addr, int mr_access_flags,
2438 struct ib_pd *pd, struct ib_udata *udata);
2439 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2440 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2442 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2443 u32 max_num_data_sg,
2444 u32 max_num_meta_sg);
2445 int (*advise_mr)(struct ib_pd *pd,
2446 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2447 struct ib_sge *sg_list, u32 num_sge,
2448 struct uverbs_attr_bundle *attrs);
2449 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2450 unsigned int *sg_offset);
2451 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2452 struct ib_mr_status *mr_status);
2453 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2454 int (*dealloc_mw)(struct ib_mw *mw);
2455 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2456 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2457 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2458 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2459 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2460 struct ib_flow_attr *flow_attr,
2461 struct ib_udata *udata);
2462 int (*destroy_flow)(struct ib_flow *flow_id);
2463 struct ib_flow_action *(*create_flow_action_esp)(
2464 struct ib_device *device,
2465 const struct ib_flow_action_attrs_esp *attr,
2466 struct uverbs_attr_bundle *attrs);
2467 int (*destroy_flow_action)(struct ib_flow_action *action);
2468 int (*modify_flow_action_esp)(
2469 struct ib_flow_action *action,
2470 const struct ib_flow_action_attrs_esp *attr,
2471 struct uverbs_attr_bundle *attrs);
2472 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2474 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2475 struct ifla_vf_info *ivf);
2476 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2477 struct ifla_vf_stats *stats);
2478 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2479 struct ifla_vf_guid *node_guid,
2480 struct ifla_vf_guid *port_guid);
2481 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2483 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2484 struct ib_wq_init_attr *init_attr,
2485 struct ib_udata *udata);
2486 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2487 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2488 u32 wq_attr_mask, struct ib_udata *udata);
2489 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2490 struct ib_rwq_ind_table_init_attr *init_attr,
2491 struct ib_udata *udata);
2492 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2493 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2494 struct ib_ucontext *context,
2495 struct ib_dm_alloc_attr *attr,
2496 struct uverbs_attr_bundle *attrs);
2497 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2498 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2499 struct ib_dm_mr_attr *attr,
2500 struct uverbs_attr_bundle *attrs);
2501 int (*create_counters)(struct ib_counters *counters,
2502 struct uverbs_attr_bundle *attrs);
2503 int (*destroy_counters)(struct ib_counters *counters);
2504 int (*read_counters)(struct ib_counters *counters,
2505 struct ib_counters_read_attr *counters_read_attr,
2506 struct uverbs_attr_bundle *attrs);
2507 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2508 int data_sg_nents, unsigned int *data_sg_offset,
2509 struct scatterlist *meta_sg, int meta_sg_nents,
2510 unsigned int *meta_sg_offset);
2513 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2514 * driver initialized data. The struct is kfree()'ed by the sysfs
2515 * core when the device is removed. A lifespan of -1 in the return
2516 * struct tells the core to set a default lifespan.
2518 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2521 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2522 * @index - The index in the value array we wish to have updated, or
2523 * num_counters if we want all stats updated
2525 * < 0 - Error, no counters updated
2526 * index - Updated the single counter pointed to by index
2527 * num_counters - Updated all counters (will reset the timestamp
2528 * and prevent further calls for lifespan milliseconds)
2529 * Drivers are allowed to update all counters in leiu of just the
2530 * one given in index at their option
2532 int (*get_hw_stats)(struct ib_device *device,
2533 struct rdma_hw_stats *stats, u8 port, int index);
2535 * This function is called once for each port when a ib device is
2538 int (*init_port)(struct ib_device *device, u8 port_num,
2539 struct kobject *port_sysfs);
2541 * Allows rdma drivers to add their own restrack attributes.
2543 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2544 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2545 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2546 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2547 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2548 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2549 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2551 /* Device lifecycle callbacks */
2553 * Called after the device becomes registered, before clients are
2556 int (*enable_driver)(struct ib_device *dev);
2558 * This is called as part of ib_dealloc_device().
2560 void (*dealloc_driver)(struct ib_device *dev);
2562 /* iWarp CM callbacks */
2563 void (*iw_add_ref)(struct ib_qp *qp);
2564 void (*iw_rem_ref)(struct ib_qp *qp);
2565 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2566 int (*iw_connect)(struct iw_cm_id *cm_id,
2567 struct iw_cm_conn_param *conn_param);
2568 int (*iw_accept)(struct iw_cm_id *cm_id,
2569 struct iw_cm_conn_param *conn_param);
2570 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2572 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2573 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2575 * counter_bind_qp - Bind a QP to a counter.
2576 * @counter - The counter to be bound. If counter->id is zero then
2577 * the driver needs to allocate a new counter and set counter->id
2579 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2581 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2582 * counter and bind it onto the default one
2584 int (*counter_unbind_qp)(struct ib_qp *qp);
2586 * counter_dealloc -De-allocate the hw counter
2588 int (*counter_dealloc)(struct rdma_counter *counter);
2590 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2591 * the driver initialized data.
2593 struct rdma_hw_stats *(*counter_alloc_stats)(
2594 struct rdma_counter *counter);
2596 * counter_update_stats - Query the stats value of this counter
2598 int (*counter_update_stats)(struct rdma_counter *counter);
2601 * Allows rdma drivers to add their own restrack attributes
2602 * dumped via 'rdma stat' iproute2 command.
2604 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2606 /* query driver for its ucontext properties */
2607 int (*query_ucontext)(struct ib_ucontext *context,
2608 struct uverbs_attr_bundle *attrs);
2610 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2611 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2612 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2613 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2614 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2615 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2616 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2617 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2618 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2621 struct ib_core_device {
2622 /* device must be the first element in structure until,
2623 * union of ib_core_device and device exists in ib_device.
2626 possible_net_t rdma_net;
2627 struct kobject *ports_kobj;
2628 struct list_head port_list;
2629 struct ib_device *owner; /* reach back to owner ib_device */
2632 struct rdma_restrack_root;
2634 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2635 struct device *dma_device;
2636 struct ib_device_ops ops;
2637 char name[IB_DEVICE_NAME_MAX];
2638 struct rcu_head rcu_head;
2640 struct list_head event_handler_list;
2641 /* Protects event_handler_list */
2642 struct rw_semaphore event_handler_rwsem;
2644 /* Protects QP's event_handler calls and open_qp list */
2645 spinlock_t qp_open_list_lock;
2647 struct rw_semaphore client_data_rwsem;
2648 struct xarray client_data;
2649 struct mutex unregistration_lock;
2651 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2652 rwlock_t cache_lock;
2654 * port_data is indexed by port number
2656 struct ib_port_data *port_data;
2658 int num_comp_vectors;
2662 struct ib_core_device coredev;
2665 /* First group for device attributes,
2666 * Second group for driver provided attributes (optional).
2667 * It is NULL terminated array.
2669 const struct attribute_group *groups[3];
2671 u64 uverbs_cmd_mask;
2673 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2677 /* Indicates kernel verbs support, should not be used in drivers */
2678 u16 kverbs_provider:1;
2679 /* CQ adaptive moderation (RDMA DIM) */
2683 struct ib_device_attr attrs;
2684 struct attribute_group *hw_stats_ag;
2685 struct rdma_hw_stats *hw_stats;
2687 #ifdef CONFIG_CGROUP_RDMA
2688 struct rdmacg_device cg_device;
2693 spinlock_t cq_pools_lock;
2694 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2696 struct rdma_restrack_root *res;
2698 const struct uapi_definition *driver_def;
2701 * Positive refcount indicates that the device is currently
2702 * registered and cannot be unregistered.
2704 refcount_t refcount;
2705 struct completion unreg_completion;
2706 struct work_struct unregistration_work;
2708 const struct rdma_link_ops *link_ops;
2710 /* Protects compat_devs xarray modifications */
2711 struct mutex compat_devs_mutex;
2712 /* Maintains compat devices for each net namespace */
2713 struct xarray compat_devs;
2715 /* Used by iWarp CM */
2716 char iw_ifname[IFNAMSIZ];
2717 u32 iw_driver_flags;
2721 struct ib_client_nl_info;
2724 int (*add)(struct ib_device *ibdev);
2725 void (*remove)(struct ib_device *, void *client_data);
2726 void (*rename)(struct ib_device *dev, void *client_data);
2727 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2728 struct ib_client_nl_info *res);
2729 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2731 /* Returns the net_dev belonging to this ib_client and matching the
2733 * @dev: An RDMA device that the net_dev use for communication.
2734 * @port: A physical port number on the RDMA device.
2735 * @pkey: P_Key that the net_dev uses if applicable.
2736 * @gid: A GID that the net_dev uses to communicate.
2737 * @addr: An IP address the net_dev is configured with.
2738 * @client_data: The device's client data set by ib_set_client_data().
2740 * An ib_client that implements a net_dev on top of RDMA devices
2741 * (such as IP over IB) should implement this callback, allowing the
2742 * rdma_cm module to find the right net_dev for a given request.
2744 * The caller is responsible for calling dev_put on the returned
2746 struct net_device *(*get_net_dev_by_params)(
2747 struct ib_device *dev,
2750 const union ib_gid *gid,
2751 const struct sockaddr *addr,
2755 struct completion uses_zero;
2758 /* kverbs are not required by the client */
2763 * IB block DMA iterator
2765 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2766 * to a HW supported page size.
2768 struct ib_block_iter {
2769 /* internal states */
2770 struct scatterlist *__sg; /* sg holding the current aligned block */
2771 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2772 unsigned int __sg_nents; /* number of SG entries */
2773 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2774 unsigned int __pg_bit; /* alignment of current block */
2777 struct ib_device *_ib_alloc_device(size_t size);
2778 #define ib_alloc_device(drv_struct, member) \
2779 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2780 BUILD_BUG_ON_ZERO(offsetof( \
2781 struct drv_struct, member))), \
2782 struct drv_struct, member)
2784 void ib_dealloc_device(struct ib_device *device);
2786 void ib_get_device_fw_str(struct ib_device *device, char *str);
2788 int ib_register_device(struct ib_device *device, const char *name,
2789 struct device *dma_device);
2790 void ib_unregister_device(struct ib_device *device);
2791 void ib_unregister_driver(enum rdma_driver_id driver_id);
2792 void ib_unregister_device_and_put(struct ib_device *device);
2793 void ib_unregister_device_queued(struct ib_device *ib_dev);
2795 int ib_register_client (struct ib_client *client);
2796 void ib_unregister_client(struct ib_client *client);
2798 void __rdma_block_iter_start(struct ib_block_iter *biter,
2799 struct scatterlist *sglist,
2801 unsigned long pgsz);
2802 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2805 * rdma_block_iter_dma_address - get the aligned dma address of the current
2806 * block held by the block iterator.
2807 * @biter: block iterator holding the memory block
2809 static inline dma_addr_t
2810 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2812 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2816 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2817 * @sglist: sglist to iterate over
2818 * @biter: block iterator holding the memory block
2819 * @nents: maximum number of sg entries to iterate over
2820 * @pgsz: best HW supported page size to use
2822 * Callers may use rdma_block_iter_dma_address() to get each
2823 * blocks aligned DMA address.
2825 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2826 for (__rdma_block_iter_start(biter, sglist, nents, \
2828 __rdma_block_iter_next(biter);)
2831 * ib_get_client_data - Get IB client context
2832 * @device:Device to get context for
2833 * @client:Client to get context for
2835 * ib_get_client_data() returns the client context data set with
2836 * ib_set_client_data(). This can only be called while the client is
2837 * registered to the device, once the ib_client remove() callback returns this
2840 static inline void *ib_get_client_data(struct ib_device *device,
2841 struct ib_client *client)
2843 return xa_load(&device->client_data, client->client_id);
2845 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2847 void ib_set_device_ops(struct ib_device *device,
2848 const struct ib_device_ops *ops);
2850 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2851 unsigned long pfn, unsigned long size, pgprot_t prot,
2852 struct rdma_user_mmap_entry *entry);
2853 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2854 struct rdma_user_mmap_entry *entry,
2856 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2857 struct rdma_user_mmap_entry *entry,
2858 size_t length, u32 min_pgoff,
2861 struct rdma_user_mmap_entry *
2862 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2863 unsigned long pgoff);
2864 struct rdma_user_mmap_entry *
2865 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2866 struct vm_area_struct *vma);
2867 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2869 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2871 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2873 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2876 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2878 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2881 static inline bool ib_is_buffer_cleared(const void __user *p,
2887 if (len > USHRT_MAX)
2890 buf = memdup_user(p, len);
2894 ret = !memchr_inv(buf, 0, len);
2899 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2903 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2907 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2908 * contains all required attributes and no attributes not allowed for
2909 * the given QP state transition.
2910 * @cur_state: Current QP state
2911 * @next_state: Next QP state
2913 * @mask: Mask of supplied QP attributes
2915 * This function is a helper function that a low-level driver's
2916 * modify_qp method can use to validate the consumer's input. It
2917 * checks that cur_state and next_state are valid QP states, that a
2918 * transition from cur_state to next_state is allowed by the IB spec,
2919 * and that the attribute mask supplied is allowed for the transition.
2921 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2922 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2924 void ib_register_event_handler(struct ib_event_handler *event_handler);
2925 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2926 void ib_dispatch_event(const struct ib_event *event);
2928 int ib_query_port(struct ib_device *device,
2929 u8 port_num, struct ib_port_attr *port_attr);
2931 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2935 * rdma_cap_ib_switch - Check if the device is IB switch
2936 * @device: Device to check
2938 * Device driver is responsible for setting is_switch bit on
2939 * in ib_device structure at init time.
2941 * Return: true if the device is IB switch.
2943 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2945 return device->is_switch;
2949 * rdma_start_port - Return the first valid port number for the device
2952 * @device: Device to be checked
2954 * Return start port number
2956 static inline u8 rdma_start_port(const struct ib_device *device)
2958 return rdma_cap_ib_switch(device) ? 0 : 1;
2962 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2963 * @device - The struct ib_device * to iterate over
2964 * @iter - The unsigned int to store the port number
2966 #define rdma_for_each_port(device, iter) \
2967 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2968 unsigned int, iter))); \
2969 iter <= rdma_end_port(device); (iter)++)
2972 * rdma_end_port - Return the last valid port number for the device
2975 * @device: Device to be checked
2977 * Return last port number
2979 static inline u8 rdma_end_port(const struct ib_device *device)
2981 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2984 static inline int rdma_is_port_valid(const struct ib_device *device,
2987 return (port >= rdma_start_port(device) &&
2988 port <= rdma_end_port(device));
2991 static inline bool rdma_is_grh_required(const struct ib_device *device,
2994 return device->port_data[port_num].immutable.core_cap_flags &
2995 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2998 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3000 return device->port_data[port_num].immutable.core_cap_flags &
3001 RDMA_CORE_CAP_PROT_IB;
3004 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3006 return device->port_data[port_num].immutable.core_cap_flags &
3007 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3010 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3012 return device->port_data[port_num].immutable.core_cap_flags &
3013 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3016 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3018 return device->port_data[port_num].immutable.core_cap_flags &
3019 RDMA_CORE_CAP_PROT_ROCE;
3022 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3024 return device->port_data[port_num].immutable.core_cap_flags &
3025 RDMA_CORE_CAP_PROT_IWARP;
3028 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3030 return rdma_protocol_ib(device, port_num) ||
3031 rdma_protocol_roce(device, port_num);
3034 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3036 return device->port_data[port_num].immutable.core_cap_flags &
3037 RDMA_CORE_CAP_PROT_RAW_PACKET;
3040 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3042 return device->port_data[port_num].immutable.core_cap_flags &
3043 RDMA_CORE_CAP_PROT_USNIC;
3047 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3048 * Management Datagrams.
3049 * @device: Device to check
3050 * @port_num: Port number to check
3052 * Management Datagrams (MAD) are a required part of the InfiniBand
3053 * specification and are supported on all InfiniBand devices. A slightly
3054 * extended version are also supported on OPA interfaces.
3056 * Return: true if the port supports sending/receiving of MAD packets.
3058 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3060 return device->port_data[port_num].immutable.core_cap_flags &
3061 RDMA_CORE_CAP_IB_MAD;
3065 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3066 * Management Datagrams.
3067 * @device: Device to check
3068 * @port_num: Port number to check
3070 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3071 * datagrams with their own versions. These OPA MADs share many but not all of
3072 * the characteristics of InfiniBand MADs.
3074 * OPA MADs differ in the following ways:
3076 * 1) MADs are variable size up to 2K
3077 * IBTA defined MADs remain fixed at 256 bytes
3078 * 2) OPA SMPs must carry valid PKeys
3079 * 3) OPA SMP packets are a different format
3081 * Return: true if the port supports OPA MAD packet formats.
3083 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3085 return device->port_data[port_num].immutable.core_cap_flags &
3086 RDMA_CORE_CAP_OPA_MAD;
3090 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3091 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3092 * @device: Device to check
3093 * @port_num: Port number to check
3095 * Each InfiniBand node is required to provide a Subnet Management Agent
3096 * that the subnet manager can access. Prior to the fabric being fully
3097 * configured by the subnet manager, the SMA is accessed via a well known
3098 * interface called the Subnet Management Interface (SMI). This interface
3099 * uses directed route packets to communicate with the SM to get around the
3100 * chicken and egg problem of the SM needing to know what's on the fabric
3101 * in order to configure the fabric, and needing to configure the fabric in
3102 * order to send packets to the devices on the fabric. These directed
3103 * route packets do not need the fabric fully configured in order to reach
3104 * their destination. The SMI is the only method allowed to send
3105 * directed route packets on an InfiniBand fabric.
3107 * Return: true if the port provides an SMI.
3109 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3111 return device->port_data[port_num].immutable.core_cap_flags &
3112 RDMA_CORE_CAP_IB_SMI;
3116 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3117 * Communication Manager.
3118 * @device: Device to check
3119 * @port_num: Port number to check
3121 * The InfiniBand Communication Manager is one of many pre-defined General
3122 * Service Agents (GSA) that are accessed via the General Service
3123 * Interface (GSI). It's role is to facilitate establishment of connections
3124 * between nodes as well as other management related tasks for established
3127 * Return: true if the port supports an IB CM (this does not guarantee that
3128 * a CM is actually running however).
3130 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3132 return device->port_data[port_num].immutable.core_cap_flags &
3133 RDMA_CORE_CAP_IB_CM;
3137 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3138 * Communication Manager.
3139 * @device: Device to check
3140 * @port_num: Port number to check
3142 * Similar to above, but specific to iWARP connections which have a different
3143 * managment protocol than InfiniBand.
3145 * Return: true if the port supports an iWARP CM (this does not guarantee that
3146 * a CM is actually running however).
3148 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3150 return device->port_data[port_num].immutable.core_cap_flags &
3151 RDMA_CORE_CAP_IW_CM;
3155 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3156 * Subnet Administration.
3157 * @device: Device to check
3158 * @port_num: Port number to check
3160 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3161 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3162 * fabrics, devices should resolve routes to other hosts by contacting the
3163 * SA to query the proper route.
3165 * Return: true if the port should act as a client to the fabric Subnet
3166 * Administration interface. This does not imply that the SA service is
3169 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3171 return device->port_data[port_num].immutable.core_cap_flags &
3172 RDMA_CORE_CAP_IB_SA;
3176 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3178 * @device: Device to check
3179 * @port_num: Port number to check
3181 * InfiniBand multicast registration is more complex than normal IPv4 or
3182 * IPv6 multicast registration. Each Host Channel Adapter must register
3183 * with the Subnet Manager when it wishes to join a multicast group. It
3184 * should do so only once regardless of how many queue pairs it subscribes
3185 * to this group. And it should leave the group only after all queue pairs
3186 * attached to the group have been detached.
3188 * Return: true if the port must undertake the additional adminstrative
3189 * overhead of registering/unregistering with the SM and tracking of the
3190 * total number of queue pairs attached to the multicast group.
3192 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3194 return rdma_cap_ib_sa(device, port_num);
3198 * rdma_cap_af_ib - Check if the port of device has the capability
3199 * Native Infiniband Address.
3200 * @device: Device to check
3201 * @port_num: Port number to check
3203 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3204 * GID. RoCE uses a different mechanism, but still generates a GID via
3205 * a prescribed mechanism and port specific data.
3207 * Return: true if the port uses a GID address to identify devices on the
3210 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3212 return device->port_data[port_num].immutable.core_cap_flags &
3213 RDMA_CORE_CAP_AF_IB;
3217 * rdma_cap_eth_ah - Check if the port of device has the capability
3218 * Ethernet Address Handle.
3219 * @device: Device to check
3220 * @port_num: Port number to check
3222 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3223 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3224 * port. Normally, packet headers are generated by the sending host
3225 * adapter, but when sending connectionless datagrams, we must manually
3226 * inject the proper headers for the fabric we are communicating over.
3228 * Return: true if we are running as a RoCE port and must force the
3229 * addition of a Global Route Header built from our Ethernet Address
3230 * Handle into our header list for connectionless packets.
3232 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3234 return device->port_data[port_num].immutable.core_cap_flags &
3235 RDMA_CORE_CAP_ETH_AH;
3239 * rdma_cap_opa_ah - Check if the port of device supports
3240 * OPA Address handles
3241 * @device: Device to check
3242 * @port_num: Port number to check
3244 * Return: true if we are running on an OPA device which supports
3245 * the extended OPA addressing.
3247 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3249 return (device->port_data[port_num].immutable.core_cap_flags &
3250 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3254 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3257 * @port_num: Port number
3259 * This MAD size includes the MAD headers and MAD payload. No other headers
3262 * Return the max MAD size required by the Port. Will return 0 if the port
3263 * does not support MADs
3265 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3267 return device->port_data[port_num].immutable.max_mad_size;
3271 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3272 * @device: Device to check
3273 * @port_num: Port number to check
3275 * RoCE GID table mechanism manages the various GIDs for a device.
3277 * NOTE: if allocating the port's GID table has failed, this call will still
3278 * return true, but any RoCE GID table API will fail.
3280 * Return: true if the port uses RoCE GID table mechanism in order to manage
3283 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3286 return rdma_protocol_roce(device, port_num) &&
3287 device->ops.add_gid && device->ops.del_gid;
3291 * Check if the device supports READ W/ INVALIDATE.
3293 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3296 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3297 * has support for it yet.
3299 return rdma_protocol_iwarp(dev, port_num);
3303 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3305 * @port_num: 1 based Port number
3307 * Return true if port is an Intel OPA port , false if not
3309 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3312 return (device->port_data[port_num].immutable.core_cap_flags &
3313 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3317 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3319 * @port_num: Port number
3320 * @mtu: enum value of MTU
3322 * Return the MTU size supported by the port as an integer value. Will return
3323 * -1 if enum value of mtu is not supported.
3325 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3328 if (rdma_core_cap_opa_port(device, port))
3329 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3331 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3335 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3337 * @port_num: Port number
3338 * @attr: port attribute
3340 * Return the MTU size supported by the port as an integer value.
3342 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3343 struct ib_port_attr *attr)
3345 if (rdma_core_cap_opa_port(device, port))
3346 return attr->phys_mtu;
3348 return ib_mtu_enum_to_int(attr->max_mtu);
3351 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3353 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3354 struct ifla_vf_info *info);
3355 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3356 struct ifla_vf_stats *stats);
3357 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3358 struct ifla_vf_guid *node_guid,
3359 struct ifla_vf_guid *port_guid);
3360 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3363 int ib_query_pkey(struct ib_device *device,
3364 u8 port_num, u16 index, u16 *pkey);
3366 int ib_modify_device(struct ib_device *device,
3367 int device_modify_mask,
3368 struct ib_device_modify *device_modify);
3370 int ib_modify_port(struct ib_device *device,
3371 u8 port_num, int port_modify_mask,
3372 struct ib_port_modify *port_modify);
3374 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3375 u8 *port_num, u16 *index);
3377 int ib_find_pkey(struct ib_device *device,
3378 u8 port_num, u16 pkey, u16 *index);
3382 * Create a memory registration for all memory in the system and place
3383 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3384 * ULPs to avoid the overhead of dynamic MRs.
3386 * This flag is generally considered unsafe and must only be used in
3387 * extremly trusted environments. Every use of it will log a warning
3388 * in the kernel log.
3390 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3393 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3394 const char *caller);
3396 #define ib_alloc_pd(device, flags) \
3397 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3399 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3402 * ib_dealloc_pd - Deallocate kernel PD
3403 * @pd: The protection domain
3405 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3407 static inline void ib_dealloc_pd(struct ib_pd *pd)
3409 int ret = ib_dealloc_pd_user(pd, NULL);
3411 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3414 enum rdma_create_ah_flags {
3415 /* In a sleepable context */
3416 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3420 * rdma_create_ah - Creates an address handle for the given address vector.
3421 * @pd: The protection domain associated with the address handle.
3422 * @ah_attr: The attributes of the address vector.
3423 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3425 * The address handle is used to reference a local or global destination
3426 * in all UD QP post sends.
3428 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3432 * rdma_create_user_ah - Creates an address handle for the given address vector.
3433 * It resolves destination mac address for ah attribute of RoCE type.
3434 * @pd: The protection domain associated with the address handle.
3435 * @ah_attr: The attributes of the address vector.
3436 * @udata: pointer to user's input output buffer information need by
3439 * It returns 0 on success and returns appropriate error code on error.
3440 * The address handle is used to reference a local or global destination
3441 * in all UD QP post sends.
3443 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3444 struct rdma_ah_attr *ah_attr,
3445 struct ib_udata *udata);
3447 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3449 * @hdr: the L3 header to parse
3450 * @net_type: type of header to parse
3451 * @sgid: place to store source gid
3452 * @dgid: place to store destination gid
3454 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3455 enum rdma_network_type net_type,
3456 union ib_gid *sgid, union ib_gid *dgid);
3459 * ib_get_rdma_header_version - Get the header version
3460 * @hdr: the L3 header to parse
3462 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3465 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3467 * @device: Device on which the received message arrived.
3468 * @port_num: Port on which the received message arrived.
3469 * @wc: Work completion associated with the received message.
3470 * @grh: References the received global route header. This parameter is
3471 * ignored unless the work completion indicates that the GRH is valid.
3472 * @ah_attr: Returned attributes that can be used when creating an address
3473 * handle for replying to the message.
3474 * When ib_init_ah_attr_from_wc() returns success,
3475 * (a) for IB link layer it optionally contains a reference to SGID attribute
3476 * when GRH is present for IB link layer.
3477 * (b) for RoCE link layer it contains a reference to SGID attribute.
3478 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3479 * attributes which are initialized using ib_init_ah_attr_from_wc().
3482 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3483 const struct ib_wc *wc, const struct ib_grh *grh,
3484 struct rdma_ah_attr *ah_attr);
3487 * ib_create_ah_from_wc - Creates an address handle associated with the
3488 * sender of the specified work completion.
3489 * @pd: The protection domain associated with the address handle.
3490 * @wc: Work completion information associated with a received message.
3491 * @grh: References the received global route header. This parameter is
3492 * ignored unless the work completion indicates that the GRH is valid.
3493 * @port_num: The outbound port number to associate with the address.
3495 * The address handle is used to reference a local or global destination
3496 * in all UD QP post sends.
3498 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3499 const struct ib_grh *grh, u8 port_num);
3502 * rdma_modify_ah - Modifies the address vector associated with an address
3504 * @ah: The address handle to modify.
3505 * @ah_attr: The new address vector attributes to associate with the
3508 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3511 * rdma_query_ah - Queries the address vector associated with an address
3513 * @ah: The address handle to query.
3514 * @ah_attr: The address vector attributes associated with the address
3517 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3519 enum rdma_destroy_ah_flags {
3520 /* In a sleepable context */
3521 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3525 * rdma_destroy_ah_user - Destroys an address handle.
3526 * @ah: The address handle to destroy.
3527 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3528 * @udata: Valid user data or NULL for kernel objects
3530 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3533 * rdma_destroy_ah - Destroys an kernel address handle.
3534 * @ah: The address handle to destroy.
3535 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3537 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3539 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3541 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3543 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3546 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3547 struct ib_srq_init_attr *srq_init_attr,
3548 struct ib_usrq_object *uobject,
3549 struct ib_udata *udata);
3550 static inline struct ib_srq *
3551 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3553 if (!pd->device->ops.create_srq)
3554 return ERR_PTR(-EOPNOTSUPP);
3556 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3560 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3561 * @srq: The SRQ to modify.
3562 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3563 * the current values of selected SRQ attributes are returned.
3564 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3565 * are being modified.
3567 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3568 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3569 * the number of receives queued drops below the limit.
3571 int ib_modify_srq(struct ib_srq *srq,
3572 struct ib_srq_attr *srq_attr,
3573 enum ib_srq_attr_mask srq_attr_mask);
3576 * ib_query_srq - Returns the attribute list and current values for the
3578 * @srq: The SRQ to query.
3579 * @srq_attr: The attributes of the specified SRQ.
3581 int ib_query_srq(struct ib_srq *srq,
3582 struct ib_srq_attr *srq_attr);
3585 * ib_destroy_srq_user - Destroys the specified SRQ.
3586 * @srq: The SRQ to destroy.
3587 * @udata: Valid user data or NULL for kernel objects
3589 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3592 * ib_destroy_srq - Destroys the specified kernel SRQ.
3593 * @srq: The SRQ to destroy.
3595 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3597 static inline void ib_destroy_srq(struct ib_srq *srq)
3599 int ret = ib_destroy_srq_user(srq, NULL);
3601 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3605 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3606 * @srq: The SRQ to post the work request on.
3607 * @recv_wr: A list of work requests to post on the receive queue.
3608 * @bad_recv_wr: On an immediate failure, this parameter will reference
3609 * the work request that failed to be posted on the QP.
3611 static inline int ib_post_srq_recv(struct ib_srq *srq,
3612 const struct ib_recv_wr *recv_wr,
3613 const struct ib_recv_wr **bad_recv_wr)
3615 const struct ib_recv_wr *dummy;
3617 return srq->device->ops.post_srq_recv(srq, recv_wr,
3618 bad_recv_wr ? : &dummy);
3621 struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
3622 struct ib_qp_init_attr *qp_init_attr,
3623 const char *caller);
3624 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3625 struct ib_qp_init_attr *init_attr)
3627 return ib_create_named_qp(pd, init_attr, KBUILD_MODNAME);
3631 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3632 * @qp: The QP to modify.
3633 * @attr: On input, specifies the QP attributes to modify. On output,
3634 * the current values of selected QP attributes are returned.
3635 * @attr_mask: A bit-mask used to specify which attributes of the QP
3636 * are being modified.
3637 * @udata: pointer to user's input output buffer information
3638 * are being modified.
3639 * It returns 0 on success and returns appropriate error code on error.
3641 int ib_modify_qp_with_udata(struct ib_qp *qp,
3642 struct ib_qp_attr *attr,
3644 struct ib_udata *udata);
3647 * ib_modify_qp - Modifies the attributes for the specified QP and then
3648 * transitions the QP to the given state.
3649 * @qp: The QP to modify.
3650 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3651 * the current values of selected QP attributes are returned.
3652 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3653 * are being modified.
3655 int ib_modify_qp(struct ib_qp *qp,
3656 struct ib_qp_attr *qp_attr,
3660 * ib_query_qp - Returns the attribute list and current values for the
3662 * @qp: The QP to query.
3663 * @qp_attr: The attributes of the specified QP.
3664 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3665 * @qp_init_attr: Additional attributes of the selected QP.
3667 * The qp_attr_mask may be used to limit the query to gathering only the
3668 * selected attributes.
3670 int ib_query_qp(struct ib_qp *qp,
3671 struct ib_qp_attr *qp_attr,
3673 struct ib_qp_init_attr *qp_init_attr);
3676 * ib_destroy_qp - Destroys the specified QP.
3677 * @qp: The QP to destroy.
3678 * @udata: Valid udata or NULL for kernel objects
3680 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3683 * ib_destroy_qp - Destroys the specified kernel QP.
3684 * @qp: The QP to destroy.
3686 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3688 static inline int ib_destroy_qp(struct ib_qp *qp)
3690 return ib_destroy_qp_user(qp, NULL);
3694 * ib_open_qp - Obtain a reference to an existing sharable QP.
3695 * @xrcd - XRC domain
3696 * @qp_open_attr: Attributes identifying the QP to open.
3698 * Returns a reference to a sharable QP.
3700 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3701 struct ib_qp_open_attr *qp_open_attr);
3704 * ib_close_qp - Release an external reference to a QP.
3705 * @qp: The QP handle to release
3707 * The opened QP handle is released by the caller. The underlying
3708 * shared QP is not destroyed until all internal references are released.
3710 int ib_close_qp(struct ib_qp *qp);
3713 * ib_post_send - Posts a list of work requests to the send queue of
3715 * @qp: The QP to post the work request on.
3716 * @send_wr: A list of work requests to post on the send queue.
3717 * @bad_send_wr: On an immediate failure, this parameter will reference
3718 * the work request that failed to be posted on the QP.
3720 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3721 * error is returned, the QP state shall not be affected,
3722 * ib_post_send() will return an immediate error after queueing any
3723 * earlier work requests in the list.
3725 static inline int ib_post_send(struct ib_qp *qp,
3726 const struct ib_send_wr *send_wr,
3727 const struct ib_send_wr **bad_send_wr)
3729 const struct ib_send_wr *dummy;
3731 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3735 * ib_post_recv - Posts a list of work requests to the receive queue of
3737 * @qp: The QP to post the work request on.
3738 * @recv_wr: A list of work requests to post on the receive queue.
3739 * @bad_recv_wr: On an immediate failure, this parameter will reference
3740 * the work request that failed to be posted on the QP.
3742 static inline int ib_post_recv(struct ib_qp *qp,
3743 const struct ib_recv_wr *recv_wr,
3744 const struct ib_recv_wr **bad_recv_wr)
3746 const struct ib_recv_wr *dummy;
3748 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3751 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3752 int comp_vector, enum ib_poll_context poll_ctx,
3753 const char *caller);
3754 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3755 int nr_cqe, int comp_vector,
3756 enum ib_poll_context poll_ctx)
3758 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3762 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3763 int nr_cqe, enum ib_poll_context poll_ctx,
3764 const char *caller);
3767 * ib_alloc_cq_any: Allocate kernel CQ
3768 * @dev: The IB device
3769 * @private: Private data attached to the CQE
3770 * @nr_cqe: Number of CQEs in the CQ
3771 * @poll_ctx: Context used for polling the CQ
3773 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3774 void *private, int nr_cqe,
3775 enum ib_poll_context poll_ctx)
3777 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3781 void ib_free_cq(struct ib_cq *cq);
3782 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3785 * ib_create_cq - Creates a CQ on the specified device.
3786 * @device: The device on which to create the CQ.
3787 * @comp_handler: A user-specified callback that is invoked when a
3788 * completion event occurs on the CQ.
3789 * @event_handler: A user-specified callback that is invoked when an
3790 * asynchronous event not associated with a completion occurs on the CQ.
3791 * @cq_context: Context associated with the CQ returned to the user via
3792 * the associated completion and event handlers.
3793 * @cq_attr: The attributes the CQ should be created upon.
3795 * Users can examine the cq structure to determine the actual CQ size.
3797 struct ib_cq *__ib_create_cq(struct ib_device *device,
3798 ib_comp_handler comp_handler,
3799 void (*event_handler)(struct ib_event *, void *),
3801 const struct ib_cq_init_attr *cq_attr,
3802 const char *caller);
3803 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3804 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3807 * ib_resize_cq - Modifies the capacity of the CQ.
3808 * @cq: The CQ to resize.
3809 * @cqe: The minimum size of the CQ.
3811 * Users can examine the cq structure to determine the actual CQ size.
3813 int ib_resize_cq(struct ib_cq *cq, int cqe);
3816 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3817 * @cq: The CQ to modify.
3818 * @cq_count: number of CQEs that will trigger an event
3819 * @cq_period: max period of time in usec before triggering an event
3822 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3825 * ib_destroy_cq_user - Destroys the specified CQ.
3826 * @cq: The CQ to destroy.
3827 * @udata: Valid user data or NULL for kernel objects
3829 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3832 * ib_destroy_cq - Destroys the specified kernel CQ.
3833 * @cq: The CQ to destroy.
3835 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3837 static inline void ib_destroy_cq(struct ib_cq *cq)
3839 int ret = ib_destroy_cq_user(cq, NULL);
3841 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3845 * ib_poll_cq - poll a CQ for completion(s)
3846 * @cq:the CQ being polled
3847 * @num_entries:maximum number of completions to return
3848 * @wc:array of at least @num_entries &struct ib_wc where completions
3851 * Poll a CQ for (possibly multiple) completions. If the return value
3852 * is < 0, an error occurred. If the return value is >= 0, it is the
3853 * number of completions returned. If the return value is
3854 * non-negative and < num_entries, then the CQ was emptied.
3856 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3859 return cq->device->ops.poll_cq(cq, num_entries, wc);
3863 * ib_req_notify_cq - Request completion notification on a CQ.
3864 * @cq: The CQ to generate an event for.
3866 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3867 * to request an event on the next solicited event or next work
3868 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3869 * may also be |ed in to request a hint about missed events, as
3873 * < 0 means an error occurred while requesting notification
3874 * == 0 means notification was requested successfully, and if
3875 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3876 * were missed and it is safe to wait for another event. In
3877 * this case is it guaranteed that any work completions added
3878 * to the CQ since the last CQ poll will trigger a completion
3879 * notification event.
3880 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3881 * in. It means that the consumer must poll the CQ again to
3882 * make sure it is empty to avoid missing an event because of a
3883 * race between requesting notification and an entry being
3884 * added to the CQ. This return value means it is possible
3885 * (but not guaranteed) that a work completion has been added
3886 * to the CQ since the last poll without triggering a
3887 * completion notification event.
3889 static inline int ib_req_notify_cq(struct ib_cq *cq,
3890 enum ib_cq_notify_flags flags)
3892 return cq->device->ops.req_notify_cq(cq, flags);
3895 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3896 int comp_vector_hint,
3897 enum ib_poll_context poll_ctx);
3899 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3902 * ib_req_ncomp_notif - Request completion notification when there are
3903 * at least the specified number of unreaped completions on the CQ.
3904 * @cq: The CQ to generate an event for.
3905 * @wc_cnt: The number of unreaped completions that should be on the
3906 * CQ before an event is generated.
3908 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3910 return cq->device->ops.req_ncomp_notif ?
3911 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3916 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3917 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3918 * address into the dma address.
3920 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3922 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3926 * ib_dma_mapping_error - check a DMA addr for error
3927 * @dev: The device for which the dma_addr was created
3928 * @dma_addr: The DMA address to check
3930 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3932 if (ib_uses_virt_dma(dev))
3934 return dma_mapping_error(dev->dma_device, dma_addr);
3938 * ib_dma_map_single - Map a kernel virtual address to DMA address
3939 * @dev: The device for which the dma_addr is to be created
3940 * @cpu_addr: The kernel virtual address
3941 * @size: The size of the region in bytes
3942 * @direction: The direction of the DMA
3944 static inline u64 ib_dma_map_single(struct ib_device *dev,
3945 void *cpu_addr, size_t size,
3946 enum dma_data_direction direction)
3948 if (ib_uses_virt_dma(dev))
3949 return (uintptr_t)cpu_addr;
3950 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3954 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3955 * @dev: The device for which the DMA address was created
3956 * @addr: The DMA address
3957 * @size: The size of the region in bytes
3958 * @direction: The direction of the DMA
3960 static inline void ib_dma_unmap_single(struct ib_device *dev,
3961 u64 addr, size_t size,
3962 enum dma_data_direction direction)
3964 if (!ib_uses_virt_dma(dev))
3965 dma_unmap_single(dev->dma_device, addr, size, direction);
3969 * ib_dma_map_page - Map a physical page to DMA address
3970 * @dev: The device for which the dma_addr is to be created
3971 * @page: The page to be mapped
3972 * @offset: The offset within the page
3973 * @size: The size of the region in bytes
3974 * @direction: The direction of the DMA
3976 static inline u64 ib_dma_map_page(struct ib_device *dev,
3978 unsigned long offset,
3980 enum dma_data_direction direction)
3982 if (ib_uses_virt_dma(dev))
3983 return (uintptr_t)(page_address(page) + offset);
3984 return dma_map_page(dev->dma_device, page, offset, size, direction);
3988 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3989 * @dev: The device for which the DMA address was created
3990 * @addr: The DMA address
3991 * @size: The size of the region in bytes
3992 * @direction: The direction of the DMA
3994 static inline void ib_dma_unmap_page(struct ib_device *dev,
3995 u64 addr, size_t size,
3996 enum dma_data_direction direction)
3998 if (!ib_uses_virt_dma(dev))
3999 dma_unmap_page(dev->dma_device, addr, size, direction);
4002 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4003 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4004 struct scatterlist *sg, int nents,
4005 enum dma_data_direction direction,
4006 unsigned long dma_attrs)
4008 if (ib_uses_virt_dma(dev))
4009 return ib_dma_virt_map_sg(dev, sg, nents);
4010 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4014 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4015 struct scatterlist *sg, int nents,
4016 enum dma_data_direction direction,
4017 unsigned long dma_attrs)
4019 if (!ib_uses_virt_dma(dev))
4020 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4025 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4026 * @dev: The device for which the DMA addresses are to be created
4027 * @sg: The array of scatter/gather entries
4028 * @nents: The number of scatter/gather entries
4029 * @direction: The direction of the DMA
4031 static inline int ib_dma_map_sg(struct ib_device *dev,
4032 struct scatterlist *sg, int nents,
4033 enum dma_data_direction direction)
4035 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4039 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4040 * @dev: The device for which the DMA addresses were created
4041 * @sg: The array of scatter/gather entries
4042 * @nents: The number of scatter/gather entries
4043 * @direction: The direction of the DMA
4045 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4046 struct scatterlist *sg, int nents,
4047 enum dma_data_direction direction)
4049 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4053 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4054 * @dev: The device to query
4056 * The returned value represents a size in bytes.
4058 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4060 if (ib_uses_virt_dma(dev))
4062 return dma_get_max_seg_size(dev->dma_device);
4066 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4067 * @dev: The device for which the DMA address was created
4068 * @addr: The DMA address
4069 * @size: The size of the region in bytes
4070 * @dir: The direction of the DMA
4072 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4075 enum dma_data_direction dir)
4077 if (!ib_uses_virt_dma(dev))
4078 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4082 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4083 * @dev: The device for which the DMA address was created
4084 * @addr: The DMA address
4085 * @size: The size of the region in bytes
4086 * @dir: The direction of the DMA
4088 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4091 enum dma_data_direction dir)
4093 if (!ib_uses_virt_dma(dev))
4094 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4097 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4098 * space. This function should be called when 'current' is the owning MM.
4100 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4101 u64 virt_addr, int mr_access_flags);
4103 /* ib_advise_mr - give an advice about an address range in a memory region */
4104 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4105 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4107 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4108 * HCA translation table.
4109 * @mr: The memory region to deregister.
4110 * @udata: Valid user data or NULL for kernel object
4112 * This function can fail, if the memory region has memory windows bound to it.
4114 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4117 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4118 * HCA translation table.
4119 * @mr: The memory region to deregister.
4121 * This function can fail, if the memory region has memory windows bound to it.
4123 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4125 static inline int ib_dereg_mr(struct ib_mr *mr)
4127 return ib_dereg_mr_user(mr, NULL);
4130 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4133 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4134 u32 max_num_data_sg,
4135 u32 max_num_meta_sg);
4138 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4140 * @mr - struct ib_mr pointer to be updated.
4141 * @newkey - new key to be used.
4143 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4145 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4146 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4150 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4151 * for calculating a new rkey for type 2 memory windows.
4152 * @rkey - the rkey to increment.
4154 static inline u32 ib_inc_rkey(u32 rkey)
4156 const u32 mask = 0x000000ff;
4157 return ((rkey + 1) & mask) | (rkey & ~mask);
4161 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4162 * @qp: QP to attach to the multicast group. The QP must be type
4164 * @gid: Multicast group GID.
4165 * @lid: Multicast group LID in host byte order.
4167 * In order to send and receive multicast packets, subnet
4168 * administration must have created the multicast group and configured
4169 * the fabric appropriately. The port associated with the specified
4170 * QP must also be a member of the multicast group.
4172 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4175 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4176 * @qp: QP to detach from the multicast group.
4177 * @gid: Multicast group GID.
4178 * @lid: Multicast group LID in host byte order.
4180 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4182 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4183 struct inode *inode, struct ib_udata *udata);
4184 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4186 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4190 * Local write permission is required if remote write or
4191 * remote atomic permission is also requested.
4193 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4194 !(flags & IB_ACCESS_LOCAL_WRITE))
4197 if (flags & ~IB_ACCESS_SUPPORTED)
4200 if (flags & IB_ACCESS_ON_DEMAND &&
4201 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4206 static inline bool ib_access_writable(int access_flags)
4209 * We have writable memory backing the MR if any of the following
4210 * access flags are set. "Local write" and "remote write" obviously
4211 * require write access. "Remote atomic" can do things like fetch and
4212 * add, which will modify memory, and "MW bind" can change permissions
4213 * by binding a window.
4215 return access_flags &
4216 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4217 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4221 * ib_check_mr_status: lightweight check of MR status.
4222 * This routine may provide status checks on a selected
4223 * ib_mr. first use is for signature status check.
4225 * @mr: A memory region.
4226 * @check_mask: Bitmask of which checks to perform from
4227 * ib_mr_status_check enumeration.
4228 * @mr_status: The container of relevant status checks.
4229 * failed checks will be indicated in the status bitmask
4230 * and the relevant info shall be in the error item.
4232 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4233 struct ib_mr_status *mr_status);
4236 * ib_device_try_get: Hold a registration lock
4237 * device: The device to lock
4239 * A device under an active registration lock cannot become unregistered. It
4240 * is only possible to obtain a registration lock on a device that is fully
4241 * registered, otherwise this function returns false.
4243 * The registration lock is only necessary for actions which require the
4244 * device to still be registered. Uses that only require the device pointer to
4245 * be valid should use get_device(&ibdev->dev) to hold the memory.
4248 static inline bool ib_device_try_get(struct ib_device *dev)
4250 return refcount_inc_not_zero(&dev->refcount);
4253 void ib_device_put(struct ib_device *device);
4254 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4255 enum rdma_driver_id driver_id);
4256 struct ib_device *ib_device_get_by_name(const char *name,
4257 enum rdma_driver_id driver_id);
4258 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4259 u16 pkey, const union ib_gid *gid,
4260 const struct sockaddr *addr);
4261 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4263 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4265 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4266 struct ib_wq_init_attr *init_attr);
4267 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4268 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4271 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4272 unsigned int *sg_offset, unsigned int page_size);
4273 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4274 int data_sg_nents, unsigned int *data_sg_offset,
4275 struct scatterlist *meta_sg, int meta_sg_nents,
4276 unsigned int *meta_sg_offset, unsigned int page_size);
4279 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4280 unsigned int *sg_offset, unsigned int page_size)
4284 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4290 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4291 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4293 void ib_drain_rq(struct ib_qp *qp);
4294 void ib_drain_sq(struct ib_qp *qp);
4295 void ib_drain_qp(struct ib_qp *qp);
4297 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u16 *speed, u8 *width);
4299 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4301 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4302 return attr->roce.dmac;
4306 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4308 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4309 attr->ib.dlid = (u16)dlid;
4310 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4311 attr->opa.dlid = dlid;
4314 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4316 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4317 return attr->ib.dlid;
4318 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4319 return attr->opa.dlid;
4323 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4328 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4333 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4336 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4337 attr->ib.src_path_bits = src_path_bits;
4338 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4339 attr->opa.src_path_bits = src_path_bits;
4342 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4344 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4345 return attr->ib.src_path_bits;
4346 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4347 return attr->opa.src_path_bits;
4351 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4354 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4355 attr->opa.make_grd = make_grd;
4358 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4360 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4361 return attr->opa.make_grd;
4365 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4367 attr->port_num = port_num;
4370 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4372 return attr->port_num;
4375 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4378 attr->static_rate = static_rate;
4381 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4383 return attr->static_rate;
4386 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4387 enum ib_ah_flags flag)
4389 attr->ah_flags = flag;
4392 static inline enum ib_ah_flags
4393 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4395 return attr->ah_flags;
4398 static inline const struct ib_global_route
4399 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4404 /*To retrieve and modify the grh */
4405 static inline struct ib_global_route
4406 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4411 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4413 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4415 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4418 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4421 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4423 grh->dgid.global.subnet_prefix = prefix;
4426 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4429 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4431 grh->dgid.global.interface_id = if_id;
4434 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4435 union ib_gid *dgid, u32 flow_label,
4436 u8 sgid_index, u8 hop_limit,
4439 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4441 attr->ah_flags = IB_AH_GRH;
4444 grh->flow_label = flow_label;
4445 grh->sgid_index = sgid_index;
4446 grh->hop_limit = hop_limit;
4447 grh->traffic_class = traffic_class;
4448 grh->sgid_attr = NULL;
4451 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4452 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4453 u32 flow_label, u8 hop_limit, u8 traffic_class,
4454 const struct ib_gid_attr *sgid_attr);
4455 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4456 const struct rdma_ah_attr *src);
4457 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4458 const struct rdma_ah_attr *new);
4459 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4462 * rdma_ah_find_type - Return address handle type.
4464 * @dev: Device to be checked
4465 * @port_num: Port number
4467 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4470 if (rdma_protocol_roce(dev, port_num))
4471 return RDMA_AH_ATTR_TYPE_ROCE;
4472 if (rdma_protocol_ib(dev, port_num)) {
4473 if (rdma_cap_opa_ah(dev, port_num))
4474 return RDMA_AH_ATTR_TYPE_OPA;
4475 return RDMA_AH_ATTR_TYPE_IB;
4478 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4482 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4483 * In the current implementation the only way to get
4484 * get the 32bit lid is from other sources for OPA.
4485 * For IB, lids will always be 16bits so cast the
4486 * value accordingly.
4490 static inline u16 ib_lid_cpu16(u32 lid)
4492 WARN_ON_ONCE(lid & 0xFFFF0000);
4497 * ib_lid_be16 - Return lid in 16bit BE encoding.
4501 static inline __be16 ib_lid_be16(u32 lid)
4503 WARN_ON_ONCE(lid & 0xFFFF0000);
4504 return cpu_to_be16((u16)lid);
4508 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4510 * @device: the rdma device
4511 * @comp_vector: index of completion vector
4513 * Returns NULL on failure, otherwise a corresponding cpu map of the
4514 * completion vector (returns all-cpus map if the device driver doesn't
4515 * implement get_vector_affinity).
4517 static inline const struct cpumask *
4518 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4520 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4521 !device->ops.get_vector_affinity)
4524 return device->ops.get_vector_affinity(device, comp_vector);
4529 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4530 * and add their gids, as needed, to the relevant RoCE devices.
4532 * @device: the rdma device
4534 void rdma_roce_rescan_device(struct ib_device *ibdev);
4536 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4538 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4540 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4541 enum rdma_netdev_t type, const char *name,
4542 unsigned char name_assign_type,
4543 void (*setup)(struct net_device *));
4545 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4546 enum rdma_netdev_t type, const char *name,
4547 unsigned char name_assign_type,
4548 void (*setup)(struct net_device *),
4549 struct net_device *netdev);
4552 * rdma_set_device_sysfs_group - Set device attributes group to have
4553 * driver specific sysfs entries at
4554 * for infiniband class.
4556 * @device: device pointer for which attributes to be created
4557 * @group: Pointer to group which should be added when device
4558 * is registered with sysfs.
4559 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4560 * group per device to have sysfs attributes.
4562 * NOTE: New drivers should not make use of this API; instead new device
4563 * parameter should be exposed via netlink command. This API and mechanism
4564 * exist only for existing drivers.
4567 rdma_set_device_sysfs_group(struct ib_device *dev,
4568 const struct attribute_group *group)
4570 dev->groups[1] = group;
4574 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4576 * @device: device pointer for which ib_device pointer to retrieve
4578 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4581 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4583 struct ib_core_device *coredev =
4584 container_of(device, struct ib_core_device, dev);
4586 return coredev->owner;
4590 * ibdev_to_node - return the NUMA node for a given ib_device
4591 * @dev: device to get the NUMA node for.
4593 static inline int ibdev_to_node(struct ib_device *ibdev)
4595 struct device *parent = ibdev->dev.parent;
4598 return NUMA_NO_NODE;
4599 return dev_to_node(parent);
4603 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4604 * ib_device holder structure from device pointer.
4606 * NOTE: New drivers should not make use of this API; This API is only for
4607 * existing drivers who have exposed sysfs entries using
4608 * rdma_set_device_sysfs_group().
4610 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4611 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4613 bool rdma_dev_access_netns(const struct ib_device *device,
4614 const struct net *net);
4616 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4617 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4618 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4621 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4624 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4625 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4628 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4630 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4632 fl_low ^= fl_high >> 14;
4633 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4637 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4638 * local and remote qpn values
4640 * This function folded the multiplication results of two qpns, 24 bit each,
4641 * fields, and converts it to a 20 bit results.
4643 * This function will create symmetric flow_label value based on the local
4644 * and remote qpn values. this will allow both the requester and responder
4645 * to calculate the same flow_label for a given connection.
4647 * This helper function should be used by driver in case the upper layer
4648 * provide a zero flow_label value. This is to improve entropy of RDMA
4649 * traffic in the network.
4651 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4653 u64 v = (u64)lqpn * rqpn;
4658 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4660 #endif /* IB_VERBS_H */