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 /* If link layer is Ethernet, this is RoCE V1 */
143 IB_GID_TYPE_ROCE = 0,
144 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
148 #define ROCE_V2_UDP_DPORT 4791
150 struct net_device __rcu *ndev;
151 struct ib_device *device;
153 enum ib_gid_type gid_type;
159 /* set the local administered indication */
160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
163 enum rdma_transport_type {
165 RDMA_TRANSPORT_IWARP,
166 RDMA_TRANSPORT_USNIC,
167 RDMA_TRANSPORT_USNIC_UDP,
168 RDMA_TRANSPORT_UNSPECIFIED,
171 enum rdma_protocol_type {
175 RDMA_PROTOCOL_USNIC_UDP
178 __attribute_const__ enum rdma_transport_type
179 rdma_node_get_transport(unsigned int node_type);
181 enum rdma_network_type {
183 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
188 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
190 if (network_type == RDMA_NETWORK_IPV4 ||
191 network_type == RDMA_NETWORK_IPV6)
192 return IB_GID_TYPE_ROCE_UDP_ENCAP;
194 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
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 (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
205 return RDMA_NETWORK_IPV4;
207 return RDMA_NETWORK_IPV6;
210 enum rdma_link_layer {
211 IB_LINK_LAYER_UNSPECIFIED,
212 IB_LINK_LAYER_INFINIBAND,
213 IB_LINK_LAYER_ETHERNET,
216 enum ib_device_cap_flags {
217 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
218 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
219 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
220 IB_DEVICE_RAW_MULTI = (1 << 3),
221 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
222 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
223 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
224 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
225 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
226 /* Not in use, former INIT_TYPE = (1 << 9),*/
227 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
228 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
229 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
230 IB_DEVICE_SRQ_RESIZE = (1 << 13),
231 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
234 * This device supports a per-device lkey or stag that can be
235 * used without performing a memory registration for the local
236 * memory. Note that ULPs should never check this flag, but
237 * instead of use the local_dma_lkey flag in the ib_pd structure,
238 * which will always contain a usable lkey.
240 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
241 /* Reserved, old SEND_W_INV = (1 << 16),*/
242 IB_DEVICE_MEM_WINDOW = (1 << 17),
244 * Devices should set IB_DEVICE_UD_IP_SUM if they support
245 * insertion of UDP and TCP checksum on outgoing UD IPoIB
246 * messages and can verify the validity of checksum for
247 * incoming messages. Setting this flag implies that the
248 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
250 IB_DEVICE_UD_IP_CSUM = (1 << 18),
251 IB_DEVICE_UD_TSO = (1 << 19),
252 IB_DEVICE_XRC = (1 << 20),
255 * This device supports the IB "base memory management extension",
256 * which includes support for fast registrations (IB_WR_REG_MR,
257 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
258 * also be set by any iWarp device which must support FRs to comply
259 * to the iWarp verbs spec. iWarp devices also support the
260 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
263 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
264 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
265 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
266 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
267 IB_DEVICE_RC_IP_CSUM = (1 << 25),
268 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
269 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
271 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
272 * support execution of WQEs that involve synchronization
273 * of I/O operations with single completion queue managed
276 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
277 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
278 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
279 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
280 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
281 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
282 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
283 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
284 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
285 /* The device supports padding incoming writes to cacheline. */
286 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
287 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
296 enum ib_odp_general_cap_bits {
297 IB_ODP_SUPPORT = 1 << 0,
298 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
301 enum ib_odp_transport_cap_bits {
302 IB_ODP_SUPPORT_SEND = 1 << 0,
303 IB_ODP_SUPPORT_RECV = 1 << 1,
304 IB_ODP_SUPPORT_WRITE = 1 << 2,
305 IB_ODP_SUPPORT_READ = 1 << 3,
306 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
307 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
311 uint64_t general_caps;
313 uint32_t rc_odp_caps;
314 uint32_t uc_odp_caps;
315 uint32_t ud_odp_caps;
316 uint32_t xrc_odp_caps;
317 } per_transport_caps;
321 /* Corresponding bit will be set if qp type from
322 * 'enum ib_qp_type' is supported, e.g.
323 * supported_qpts |= 1 << IB_QPT_UD
326 u32 max_rwq_indirection_tables;
327 u32 max_rwq_indirection_table_size;
330 enum ib_tm_cap_flags {
331 /* Support tag matching with rendezvous offload for RC transport */
332 IB_TM_CAP_RNDV_RC = 1 << 0,
336 /* Max size of RNDV header */
337 u32 max_rndv_hdr_size;
338 /* Max number of entries in tag matching list */
340 /* From enum ib_tm_cap_flags */
342 /* Max number of outstanding list operations */
344 /* Max number of SGE in tag matching entry */
348 struct ib_cq_init_attr {
354 enum ib_cq_attr_mask {
355 IB_CQ_MODERATE = 1 << 0,
359 u16 max_cq_moderation_count;
360 u16 max_cq_moderation_period;
363 struct ib_dm_mr_attr {
369 struct ib_dm_alloc_attr {
375 struct ib_device_attr {
377 __be64 sys_image_guid;
385 u64 device_cap_flags;
396 int max_qp_init_rd_atom;
397 int max_ee_init_rd_atom;
398 enum ib_atomic_cap atomic_cap;
399 enum ib_atomic_cap masked_atomic_cap;
406 int max_mcast_qp_attach;
407 int max_total_mcast_qp_attach;
412 unsigned int max_fast_reg_page_list_len;
413 unsigned int max_pi_fast_reg_page_list_len;
415 u8 local_ca_ack_delay;
418 struct ib_odp_caps odp_caps;
419 uint64_t timestamp_mask;
420 uint64_t hca_core_clock; /* in KHZ */
421 struct ib_rss_caps rss_caps;
423 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
424 struct ib_tm_caps tm_caps;
425 struct ib_cq_caps cq_caps;
427 /* Max entries for sgl for optimized performance per READ */
444 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
447 case IB_MTU_256: return 256;
448 case IB_MTU_512: return 512;
449 case IB_MTU_1024: return 1024;
450 case IB_MTU_2048: return 2048;
451 case IB_MTU_4096: return 4096;
456 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
460 else if (mtu >= 2048)
462 else if (mtu >= 1024)
470 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
478 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
482 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
485 return OPA_MTU_10240;
486 else if (mtu >= 8192)
489 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
498 IB_PORT_ACTIVE_DEFER = 5
501 enum ib_port_phys_state {
502 IB_PORT_PHYS_STATE_SLEEP = 1,
503 IB_PORT_PHYS_STATE_POLLING = 2,
504 IB_PORT_PHYS_STATE_DISABLED = 3,
505 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
506 IB_PORT_PHYS_STATE_LINK_UP = 5,
507 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
508 IB_PORT_PHYS_STATE_PHY_TEST = 7,
519 static inline int ib_width_enum_to_int(enum ib_port_width width)
522 case IB_WIDTH_1X: return 1;
523 case IB_WIDTH_2X: return 2;
524 case IB_WIDTH_4X: return 4;
525 case IB_WIDTH_8X: return 8;
526 case IB_WIDTH_12X: return 12;
542 * struct rdma_hw_stats
543 * @lock - Mutex to protect parallel write access to lifespan and values
544 * of counters, which are 64bits and not guaranteeed to be written
545 * atomicaly on 32bits systems.
546 * @timestamp - Used by the core code to track when the last update was
547 * @lifespan - Used by the core code to determine how old the counters
548 * should be before being updated again. Stored in jiffies, defaults
549 * to 10 milliseconds, drivers can override the default be specifying
550 * their own value during their allocation routine.
551 * @name - Array of pointers to static names used for the counters in
553 * @num_counters - How many hardware counters there are. If name is
554 * shorter than this number, a kernel oops will result. Driver authors
555 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
556 * in their code to prevent this.
557 * @value - Array of u64 counters that are accessed by the sysfs code and
558 * filled in by the drivers get_stats routine
560 struct rdma_hw_stats {
561 struct mutex lock; /* Protect lifespan and values[] */
562 unsigned long timestamp;
563 unsigned long lifespan;
564 const char * const *names;
569 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
571 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
573 * @names - Array of static const char *
574 * @num_counters - How many elements in array
575 * @lifespan - How many milliseconds between updates
577 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
578 const char * const *names, int num_counters,
579 unsigned long lifespan)
581 struct rdma_hw_stats *stats;
583 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
587 stats->names = names;
588 stats->num_counters = num_counters;
589 stats->lifespan = msecs_to_jiffies(lifespan);
595 /* Define bits for the various functionality this port needs to be supported by
598 /* Management 0x00000FFF */
599 #define RDMA_CORE_CAP_IB_MAD 0x00000001
600 #define RDMA_CORE_CAP_IB_SMI 0x00000002
601 #define RDMA_CORE_CAP_IB_CM 0x00000004
602 #define RDMA_CORE_CAP_IW_CM 0x00000008
603 #define RDMA_CORE_CAP_IB_SA 0x00000010
604 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
606 /* Address format 0x000FF000 */
607 #define RDMA_CORE_CAP_AF_IB 0x00001000
608 #define RDMA_CORE_CAP_ETH_AH 0x00002000
609 #define RDMA_CORE_CAP_OPA_AH 0x00004000
610 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
612 /* Protocol 0xFFF00000 */
613 #define RDMA_CORE_CAP_PROT_IB 0x00100000
614 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
615 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
616 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
617 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
618 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
620 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
621 | RDMA_CORE_CAP_PROT_ROCE \
622 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
624 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
625 | RDMA_CORE_CAP_IB_MAD \
626 | RDMA_CORE_CAP_IB_SMI \
627 | RDMA_CORE_CAP_IB_CM \
628 | RDMA_CORE_CAP_IB_SA \
629 | RDMA_CORE_CAP_AF_IB)
630 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
631 | RDMA_CORE_CAP_IB_MAD \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_AF_IB \
634 | RDMA_CORE_CAP_ETH_AH)
635 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
636 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
637 | RDMA_CORE_CAP_IB_MAD \
638 | RDMA_CORE_CAP_IB_CM \
639 | RDMA_CORE_CAP_AF_IB \
640 | RDMA_CORE_CAP_ETH_AH)
641 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
642 | RDMA_CORE_CAP_IW_CM)
643 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
644 | RDMA_CORE_CAP_OPA_MAD)
646 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
648 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
650 struct ib_port_attr {
652 enum ib_port_state state;
654 enum ib_mtu active_mtu;
657 unsigned int ip_gids:1;
658 /* This is the value from PortInfo CapabilityMask, defined by IBA */
677 enum ib_device_modify_flags {
678 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
679 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
682 #define IB_DEVICE_NODE_DESC_MAX 64
684 struct ib_device_modify {
686 char node_desc[IB_DEVICE_NODE_DESC_MAX];
689 enum ib_port_modify_flags {
690 IB_PORT_SHUTDOWN = 1,
691 IB_PORT_INIT_TYPE = (1<<2),
692 IB_PORT_RESET_QKEY_CNTR = (1<<3),
693 IB_PORT_OPA_MASK_CHG = (1<<4)
696 struct ib_port_modify {
697 u32 set_port_cap_mask;
698 u32 clr_port_cap_mask;
706 IB_EVENT_QP_ACCESS_ERR,
710 IB_EVENT_PATH_MIG_ERR,
711 IB_EVENT_DEVICE_FATAL,
712 IB_EVENT_PORT_ACTIVE,
715 IB_EVENT_PKEY_CHANGE,
718 IB_EVENT_SRQ_LIMIT_REACHED,
719 IB_EVENT_QP_LAST_WQE_REACHED,
720 IB_EVENT_CLIENT_REREGISTER,
725 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
728 struct ib_device *device;
736 enum ib_event_type event;
739 struct ib_event_handler {
740 struct ib_device *device;
741 void (*handler)(struct ib_event_handler *, struct ib_event *);
742 struct list_head list;
745 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
747 (_ptr)->device = _device; \
748 (_ptr)->handler = _handler; \
749 INIT_LIST_HEAD(&(_ptr)->list); \
752 struct ib_global_route {
753 const struct ib_gid_attr *sgid_attr;
762 __be32 version_tclass_flow;
770 union rdma_network_hdr {
773 /* The IB spec states that if it's IPv4, the header
774 * is located in the last 20 bytes of the header.
777 struct iphdr roce4grh;
781 #define IB_QPN_MASK 0xFFFFFF
784 IB_MULTICAST_QPN = 0xffffff
787 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
788 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
795 IB_RATE_PORT_CURRENT = 0,
796 IB_RATE_2_5_GBPS = 2,
804 IB_RATE_120_GBPS = 10,
805 IB_RATE_14_GBPS = 11,
806 IB_RATE_56_GBPS = 12,
807 IB_RATE_112_GBPS = 13,
808 IB_RATE_168_GBPS = 14,
809 IB_RATE_25_GBPS = 15,
810 IB_RATE_100_GBPS = 16,
811 IB_RATE_200_GBPS = 17,
812 IB_RATE_300_GBPS = 18,
813 IB_RATE_28_GBPS = 19,
814 IB_RATE_50_GBPS = 20,
815 IB_RATE_400_GBPS = 21,
816 IB_RATE_600_GBPS = 22,
820 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
821 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
822 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
823 * @rate: rate to convert.
825 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
828 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
829 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
830 * @rate: rate to convert.
832 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
836 * enum ib_mr_type - memory region type
837 * @IB_MR_TYPE_MEM_REG: memory region that is used for
838 * normal registration
839 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
840 * register any arbitrary sg lists (without
841 * the normal mr constraints - see
843 * @IB_MR_TYPE_DM: memory region that is used for device
844 * memory registration
845 * @IB_MR_TYPE_USER: memory region that is used for the user-space
847 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
848 * without address translations (VA=PA)
849 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
850 * data integrity operations
858 IB_MR_TYPE_INTEGRITY,
861 enum ib_mr_status_check {
862 IB_MR_CHECK_SIG_STATUS = 1,
866 * struct ib_mr_status - Memory region status container
868 * @fail_status: Bitmask of MR checks status. For each
869 * failed check a corresponding status bit is set.
870 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
873 struct ib_mr_status {
875 struct ib_sig_err sig_err;
879 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
881 * @mult: multiple to convert.
883 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
885 struct rdma_ah_init_attr {
886 struct rdma_ah_attr *ah_attr;
888 struct net_device *xmit_slave;
891 enum rdma_ah_attr_type {
892 RDMA_AH_ATTR_TYPE_UNDEFINED,
893 RDMA_AH_ATTR_TYPE_IB,
894 RDMA_AH_ATTR_TYPE_ROCE,
895 RDMA_AH_ATTR_TYPE_OPA,
903 struct roce_ah_attr {
913 struct rdma_ah_attr {
914 struct ib_global_route grh;
919 enum rdma_ah_attr_type type;
921 struct ib_ah_attr ib;
922 struct roce_ah_attr roce;
923 struct opa_ah_attr opa;
931 IB_WC_LOC_EEC_OP_ERR,
936 IB_WC_LOC_ACCESS_ERR,
937 IB_WC_REM_INV_REQ_ERR,
938 IB_WC_REM_ACCESS_ERR,
941 IB_WC_RNR_RETRY_EXC_ERR,
942 IB_WC_LOC_RDD_VIOL_ERR,
943 IB_WC_REM_INV_RD_REQ_ERR,
946 IB_WC_INV_EEC_STATE_ERR,
948 IB_WC_RESP_TIMEOUT_ERR,
952 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
963 IB_WC_MASKED_COMP_SWAP,
964 IB_WC_MASKED_FETCH_ADD,
966 * Set value of IB_WC_RECV so consumers can test if a completion is a
967 * receive by testing (opcode & IB_WC_RECV).
970 IB_WC_RECV_RDMA_WITH_IMM
975 IB_WC_WITH_IMM = (1<<1),
976 IB_WC_WITH_INVALIDATE = (1<<2),
977 IB_WC_IP_CSUM_OK = (1<<3),
978 IB_WC_WITH_SMAC = (1<<4),
979 IB_WC_WITH_VLAN = (1<<5),
980 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
986 struct ib_cqe *wr_cqe;
988 enum ib_wc_status status;
989 enum ib_wc_opcode opcode;
1003 u8 port_num; /* valid only for DR SMPs on switches */
1006 u8 network_hdr_type;
1009 enum ib_cq_notify_flags {
1010 IB_CQ_SOLICITED = 1 << 0,
1011 IB_CQ_NEXT_COMP = 1 << 1,
1012 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1013 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1017 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1018 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1019 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1022 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1024 return srq_type == IB_SRQT_XRC ||
1025 srq_type == IB_SRQT_TM;
1028 enum ib_srq_attr_mask {
1029 IB_SRQ_MAX_WR = 1 << 0,
1030 IB_SRQ_LIMIT = 1 << 1,
1033 struct ib_srq_attr {
1039 struct ib_srq_init_attr {
1040 void (*event_handler)(struct ib_event *, void *);
1042 struct ib_srq_attr attr;
1043 enum ib_srq_type srq_type;
1049 struct ib_xrcd *xrcd;
1064 u32 max_inline_data;
1067 * Maximum number of rdma_rw_ctx structures in flight at a time.
1068 * ib_create_qp() will calculate the right amount of neededed WRs
1069 * and MRs based on this.
1081 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1082 * here (and in that order) since the MAD layer uses them as
1083 * indices into a 2-entry table.
1088 IB_QPT_RC = IB_UVERBS_QPT_RC,
1089 IB_QPT_UC = IB_UVERBS_QPT_UC,
1090 IB_QPT_UD = IB_UVERBS_QPT_UD,
1092 IB_QPT_RAW_ETHERTYPE,
1093 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1094 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1095 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1097 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1098 /* Reserve a range for qp types internal to the low level driver.
1099 * These qp types will not be visible at the IB core layer, so the
1100 * IB_QPT_MAX usages should not be affected in the core layer
1102 IB_QPT_RESERVED1 = 0x1000,
1114 enum ib_qp_create_flags {
1115 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1116 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1117 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1118 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1119 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1120 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1121 IB_QP_CREATE_NETIF_QP = 1 << 5,
1122 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1123 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1124 IB_QP_CREATE_SCATTER_FCS =
1125 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1126 IB_QP_CREATE_CVLAN_STRIPPING =
1127 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1128 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1129 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1130 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1131 /* reserve bits 26-31 for low level drivers' internal use */
1132 IB_QP_CREATE_RESERVED_START = 1 << 26,
1133 IB_QP_CREATE_RESERVED_END = 1 << 31,
1137 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1138 * callback to destroy the passed in QP.
1141 struct ib_qp_init_attr {
1142 /* Consumer's event_handler callback must not block */
1143 void (*event_handler)(struct ib_event *, void *);
1146 struct ib_cq *send_cq;
1147 struct ib_cq *recv_cq;
1149 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1150 struct ib_qp_cap cap;
1151 enum ib_sig_type sq_sig_type;
1152 enum ib_qp_type qp_type;
1156 * Only needed for special QP types, or when using the RW API.
1159 struct ib_rwq_ind_table *rwq_ind_tbl;
1163 struct ib_qp_open_attr {
1164 void (*event_handler)(struct ib_event *, void *);
1167 enum ib_qp_type qp_type;
1170 enum ib_rnr_timeout {
1171 IB_RNR_TIMER_655_36 = 0,
1172 IB_RNR_TIMER_000_01 = 1,
1173 IB_RNR_TIMER_000_02 = 2,
1174 IB_RNR_TIMER_000_03 = 3,
1175 IB_RNR_TIMER_000_04 = 4,
1176 IB_RNR_TIMER_000_06 = 5,
1177 IB_RNR_TIMER_000_08 = 6,
1178 IB_RNR_TIMER_000_12 = 7,
1179 IB_RNR_TIMER_000_16 = 8,
1180 IB_RNR_TIMER_000_24 = 9,
1181 IB_RNR_TIMER_000_32 = 10,
1182 IB_RNR_TIMER_000_48 = 11,
1183 IB_RNR_TIMER_000_64 = 12,
1184 IB_RNR_TIMER_000_96 = 13,
1185 IB_RNR_TIMER_001_28 = 14,
1186 IB_RNR_TIMER_001_92 = 15,
1187 IB_RNR_TIMER_002_56 = 16,
1188 IB_RNR_TIMER_003_84 = 17,
1189 IB_RNR_TIMER_005_12 = 18,
1190 IB_RNR_TIMER_007_68 = 19,
1191 IB_RNR_TIMER_010_24 = 20,
1192 IB_RNR_TIMER_015_36 = 21,
1193 IB_RNR_TIMER_020_48 = 22,
1194 IB_RNR_TIMER_030_72 = 23,
1195 IB_RNR_TIMER_040_96 = 24,
1196 IB_RNR_TIMER_061_44 = 25,
1197 IB_RNR_TIMER_081_92 = 26,
1198 IB_RNR_TIMER_122_88 = 27,
1199 IB_RNR_TIMER_163_84 = 28,
1200 IB_RNR_TIMER_245_76 = 29,
1201 IB_RNR_TIMER_327_68 = 30,
1202 IB_RNR_TIMER_491_52 = 31
1205 enum ib_qp_attr_mask {
1207 IB_QP_CUR_STATE = (1<<1),
1208 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1209 IB_QP_ACCESS_FLAGS = (1<<3),
1210 IB_QP_PKEY_INDEX = (1<<4),
1211 IB_QP_PORT = (1<<5),
1212 IB_QP_QKEY = (1<<6),
1214 IB_QP_PATH_MTU = (1<<8),
1215 IB_QP_TIMEOUT = (1<<9),
1216 IB_QP_RETRY_CNT = (1<<10),
1217 IB_QP_RNR_RETRY = (1<<11),
1218 IB_QP_RQ_PSN = (1<<12),
1219 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1220 IB_QP_ALT_PATH = (1<<14),
1221 IB_QP_MIN_RNR_TIMER = (1<<15),
1222 IB_QP_SQ_PSN = (1<<16),
1223 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1224 IB_QP_PATH_MIG_STATE = (1<<18),
1225 IB_QP_CAP = (1<<19),
1226 IB_QP_DEST_QPN = (1<<20),
1227 IB_QP_RESERVED1 = (1<<21),
1228 IB_QP_RESERVED2 = (1<<22),
1229 IB_QP_RESERVED3 = (1<<23),
1230 IB_QP_RESERVED4 = (1<<24),
1231 IB_QP_RATE_LIMIT = (1<<25),
1256 enum ib_qp_state qp_state;
1257 enum ib_qp_state cur_qp_state;
1258 enum ib_mtu path_mtu;
1259 enum ib_mig_state path_mig_state;
1264 int qp_access_flags;
1265 struct ib_qp_cap cap;
1266 struct rdma_ah_attr ah_attr;
1267 struct rdma_ah_attr alt_ah_attr;
1270 u8 en_sqd_async_notify;
1273 u8 max_dest_rd_atomic;
1282 struct net_device *xmit_slave;
1286 /* These are shared with userspace */
1287 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1288 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1289 IB_WR_SEND = IB_UVERBS_WR_SEND,
1290 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1291 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1292 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1293 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1294 IB_WR_LSO = IB_UVERBS_WR_TSO,
1295 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1296 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1297 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1298 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1299 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1300 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1301 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1303 /* These are kernel only and can not be issued by userspace */
1304 IB_WR_REG_MR = 0x20,
1305 IB_WR_REG_MR_INTEGRITY,
1307 /* reserve values for low level drivers' internal use.
1308 * These values will not be used at all in the ib core layer.
1310 IB_WR_RESERVED1 = 0xf0,
1322 enum ib_send_flags {
1324 IB_SEND_SIGNALED = (1<<1),
1325 IB_SEND_SOLICITED = (1<<2),
1326 IB_SEND_INLINE = (1<<3),
1327 IB_SEND_IP_CSUM = (1<<4),
1329 /* reserve bits 26-31 for low level drivers' internal use */
1330 IB_SEND_RESERVED_START = (1 << 26),
1331 IB_SEND_RESERVED_END = (1 << 31),
1341 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1345 struct ib_send_wr *next;
1348 struct ib_cqe *wr_cqe;
1350 struct ib_sge *sg_list;
1352 enum ib_wr_opcode opcode;
1356 u32 invalidate_rkey;
1361 struct ib_send_wr wr;
1366 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1368 return container_of(wr, struct ib_rdma_wr, wr);
1371 struct ib_atomic_wr {
1372 struct ib_send_wr wr;
1376 u64 compare_add_mask;
1381 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1383 return container_of(wr, struct ib_atomic_wr, wr);
1387 struct ib_send_wr wr;
1394 u16 pkey_index; /* valid for GSI only */
1395 u8 port_num; /* valid for DR SMPs on switch only */
1398 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1400 return container_of(wr, struct ib_ud_wr, wr);
1404 struct ib_send_wr wr;
1410 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1412 return container_of(wr, struct ib_reg_wr, wr);
1416 struct ib_recv_wr *next;
1419 struct ib_cqe *wr_cqe;
1421 struct ib_sge *sg_list;
1425 enum ib_access_flags {
1426 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1427 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1428 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1429 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1430 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1431 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1432 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1433 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1434 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1436 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1437 IB_ACCESS_SUPPORTED =
1438 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1442 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1443 * are hidden here instead of a uapi header!
1445 enum ib_mr_rereg_flags {
1446 IB_MR_REREG_TRANS = 1,
1447 IB_MR_REREG_PD = (1<<1),
1448 IB_MR_REREG_ACCESS = (1<<2),
1449 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1454 enum rdma_remove_reason {
1456 * Userspace requested uobject deletion or initial try
1457 * to remove uobject via cleanup. Call could fail
1459 RDMA_REMOVE_DESTROY,
1460 /* Context deletion. This call should delete the actual object itself */
1462 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1463 RDMA_REMOVE_DRIVER_REMOVE,
1464 /* uobj is being cleaned-up before being committed */
1468 struct ib_rdmacg_object {
1469 #ifdef CONFIG_CGROUP_RDMA
1470 struct rdma_cgroup *cg; /* owner rdma cgroup */
1474 struct ib_ucontext {
1475 struct ib_device *device;
1476 struct ib_uverbs_file *ufile;
1478 * 'closing' can be read by the driver only during a destroy callback,
1479 * it is set when we are closing the file descriptor and indicates
1480 * that mm_sem may be locked.
1484 bool cleanup_retryable;
1486 struct ib_rdmacg_object cg_obj;
1488 * Implementation details of the RDMA core, don't use in drivers:
1490 struct rdma_restrack_entry res;
1491 struct xarray mmap_xa;
1495 u64 user_handle; /* handle given to us by userspace */
1496 /* ufile & ucontext owning this object */
1497 struct ib_uverbs_file *ufile;
1498 /* FIXME, save memory: ufile->context == context */
1499 struct ib_ucontext *context; /* associated user context */
1500 void *object; /* containing object */
1501 struct list_head list; /* link to context's list */
1502 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1503 int id; /* index into kernel idr */
1505 atomic_t usecnt; /* protects exclusive access */
1506 struct rcu_head rcu; /* kfree_rcu() overhead */
1508 const struct uverbs_api_object *uapi_object;
1512 const void __user *inbuf;
1513 void __user *outbuf;
1521 struct ib_device *device;
1522 struct ib_uobject *uobject;
1523 atomic_t usecnt; /* count all resources */
1525 u32 unsafe_global_rkey;
1528 * Implementation details of the RDMA core, don't use in drivers:
1530 struct ib_mr *__internal_mr;
1531 struct rdma_restrack_entry res;
1535 struct ib_device *device;
1536 atomic_t usecnt; /* count all exposed resources */
1537 struct inode *inode;
1538 struct rw_semaphore tgt_qps_rwsem;
1539 struct xarray tgt_qps;
1543 struct ib_device *device;
1545 struct ib_uobject *uobject;
1546 const struct ib_gid_attr *sgid_attr;
1547 enum rdma_ah_attr_type type;
1550 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1552 enum ib_poll_context {
1553 IB_POLL_SOFTIRQ, /* poll from softirq context */
1554 IB_POLL_WORKQUEUE, /* poll from workqueue */
1555 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1556 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1558 IB_POLL_DIRECT, /* caller context, no hw completions */
1562 struct ib_device *device;
1563 struct ib_ucq_object *uobject;
1564 ib_comp_handler comp_handler;
1565 void (*event_handler)(struct ib_event *, void *);
1568 unsigned int cqe_used;
1569 atomic_t usecnt; /* count number of work queues */
1570 enum ib_poll_context poll_ctx;
1572 struct list_head pool_entry;
1574 struct irq_poll iop;
1575 struct work_struct work;
1577 struct workqueue_struct *comp_wq;
1580 /* updated only by trace points */
1584 unsigned int comp_vector;
1587 * Implementation details of the RDMA core, don't use in drivers:
1589 struct rdma_restrack_entry res;
1593 struct ib_device *device;
1595 struct ib_usrq_object *uobject;
1596 void (*event_handler)(struct ib_event *, void *);
1598 enum ib_srq_type srq_type;
1605 struct ib_xrcd *xrcd;
1612 enum ib_raw_packet_caps {
1613 /* Strip cvlan from incoming packet and report it in the matching work
1614 * completion is supported.
1616 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1617 /* Scatter FCS field of an incoming packet to host memory is supported.
1619 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1620 /* Checksum offloads are supported (for both send and receive). */
1621 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1622 /* When a packet is received for an RQ with no receive WQEs, the
1623 * packet processing is delayed.
1625 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1629 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1639 struct ib_device *device;
1640 struct ib_uwq_object *uobject;
1642 void (*event_handler)(struct ib_event *, void *);
1646 enum ib_wq_state state;
1647 enum ib_wq_type wq_type;
1652 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1653 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1654 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1655 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1656 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1659 struct ib_wq_init_attr {
1661 enum ib_wq_type wq_type;
1665 void (*event_handler)(struct ib_event *, void *);
1666 u32 create_flags; /* Use enum ib_wq_flags */
1669 enum ib_wq_attr_mask {
1670 IB_WQ_STATE = 1 << 0,
1671 IB_WQ_CUR_STATE = 1 << 1,
1672 IB_WQ_FLAGS = 1 << 2,
1676 enum ib_wq_state wq_state;
1677 enum ib_wq_state curr_wq_state;
1678 u32 flags; /* Use enum ib_wq_flags */
1679 u32 flags_mask; /* Use enum ib_wq_flags */
1682 struct ib_rwq_ind_table {
1683 struct ib_device *device;
1684 struct ib_uobject *uobject;
1687 u32 log_ind_tbl_size;
1688 struct ib_wq **ind_tbl;
1691 struct ib_rwq_ind_table_init_attr {
1692 u32 log_ind_tbl_size;
1693 /* Each entry is a pointer to Receive Work Queue */
1694 struct ib_wq **ind_tbl;
1697 enum port_pkey_state {
1698 IB_PORT_PKEY_NOT_VALID = 0,
1699 IB_PORT_PKEY_VALID = 1,
1700 IB_PORT_PKEY_LISTED = 2,
1703 struct ib_qp_security;
1705 struct ib_port_pkey {
1706 enum port_pkey_state state;
1709 struct list_head qp_list;
1710 struct list_head to_error_list;
1711 struct ib_qp_security *sec;
1714 struct ib_ports_pkeys {
1715 struct ib_port_pkey main;
1716 struct ib_port_pkey alt;
1719 struct ib_qp_security {
1721 struct ib_device *dev;
1722 /* Hold this mutex when changing port and pkey settings. */
1724 struct ib_ports_pkeys *ports_pkeys;
1725 /* A list of all open shared QP handles. Required to enforce security
1726 * properly for all users of a shared QP.
1728 struct list_head shared_qp_list;
1731 atomic_t error_list_count;
1732 struct completion error_complete;
1733 int error_comps_pending;
1737 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1738 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1741 struct ib_device *device;
1743 struct ib_cq *send_cq;
1744 struct ib_cq *recv_cq;
1747 struct list_head rdma_mrs;
1748 struct list_head sig_mrs;
1750 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1751 struct list_head xrcd_list;
1753 /* count times opened, mcast attaches, flow attaches */
1755 struct list_head open_list;
1756 struct ib_qp *real_qp;
1757 struct ib_uqp_object *uobject;
1758 void (*event_handler)(struct ib_event *, void *);
1760 /* sgid_attrs associated with the AV's */
1761 const struct ib_gid_attr *av_sgid_attr;
1762 const struct ib_gid_attr *alt_path_sgid_attr;
1766 enum ib_qp_type qp_type;
1767 struct ib_rwq_ind_table *rwq_ind_tbl;
1768 struct ib_qp_security *qp_sec;
1773 * Implementation details of the RDMA core, don't use in drivers:
1775 struct rdma_restrack_entry res;
1777 /* The counter the qp is bind to */
1778 struct rdma_counter *counter;
1782 struct ib_device *device;
1785 struct ib_uobject *uobject;
1790 struct ib_device *device;
1796 unsigned int page_size;
1797 enum ib_mr_type type;
1800 struct ib_uobject *uobject; /* user */
1801 struct list_head qp_entry; /* FR */
1805 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1807 * Implementation details of the RDMA core, don't use in drivers:
1809 struct rdma_restrack_entry res;
1813 struct ib_device *device;
1815 struct ib_uobject *uobject;
1817 enum ib_mw_type type;
1820 /* Supported steering options */
1821 enum ib_flow_attr_type {
1822 /* steering according to rule specifications */
1823 IB_FLOW_ATTR_NORMAL = 0x0,
1824 /* default unicast and multicast rule -
1825 * receive all Eth traffic which isn't steered to any QP
1827 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1828 /* default multicast rule -
1829 * receive all Eth multicast traffic which isn't steered to any QP
1831 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1832 /* sniffer rule - receive all port traffic */
1833 IB_FLOW_ATTR_SNIFFER = 0x3
1836 /* Supported steering header types */
1837 enum ib_flow_spec_type {
1839 IB_FLOW_SPEC_ETH = 0x20,
1840 IB_FLOW_SPEC_IB = 0x22,
1842 IB_FLOW_SPEC_IPV4 = 0x30,
1843 IB_FLOW_SPEC_IPV6 = 0x31,
1844 IB_FLOW_SPEC_ESP = 0x34,
1846 IB_FLOW_SPEC_TCP = 0x40,
1847 IB_FLOW_SPEC_UDP = 0x41,
1848 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1849 IB_FLOW_SPEC_GRE = 0x51,
1850 IB_FLOW_SPEC_MPLS = 0x60,
1851 IB_FLOW_SPEC_INNER = 0x100,
1853 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1854 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1855 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1856 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1858 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1859 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1861 enum ib_flow_flags {
1862 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1863 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1864 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1867 struct ib_flow_eth_filter {
1876 struct ib_flow_spec_eth {
1879 struct ib_flow_eth_filter val;
1880 struct ib_flow_eth_filter mask;
1883 struct ib_flow_ib_filter {
1890 struct ib_flow_spec_ib {
1893 struct ib_flow_ib_filter val;
1894 struct ib_flow_ib_filter mask;
1897 /* IPv4 header flags */
1898 enum ib_ipv4_flags {
1899 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1900 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1901 last have this flag set */
1904 struct ib_flow_ipv4_filter {
1915 struct ib_flow_spec_ipv4 {
1918 struct ib_flow_ipv4_filter val;
1919 struct ib_flow_ipv4_filter mask;
1922 struct ib_flow_ipv6_filter {
1933 struct ib_flow_spec_ipv6 {
1936 struct ib_flow_ipv6_filter val;
1937 struct ib_flow_ipv6_filter mask;
1940 struct ib_flow_tcp_udp_filter {
1947 struct ib_flow_spec_tcp_udp {
1950 struct ib_flow_tcp_udp_filter val;
1951 struct ib_flow_tcp_udp_filter mask;
1954 struct ib_flow_tunnel_filter {
1959 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1960 * the tunnel_id from val has the vni value
1962 struct ib_flow_spec_tunnel {
1965 struct ib_flow_tunnel_filter val;
1966 struct ib_flow_tunnel_filter mask;
1969 struct ib_flow_esp_filter {
1976 struct ib_flow_spec_esp {
1979 struct ib_flow_esp_filter val;
1980 struct ib_flow_esp_filter mask;
1983 struct ib_flow_gre_filter {
1984 __be16 c_ks_res0_ver;
1991 struct ib_flow_spec_gre {
1994 struct ib_flow_gre_filter val;
1995 struct ib_flow_gre_filter mask;
1998 struct ib_flow_mpls_filter {
2004 struct ib_flow_spec_mpls {
2007 struct ib_flow_mpls_filter val;
2008 struct ib_flow_mpls_filter mask;
2011 struct ib_flow_spec_action_tag {
2012 enum ib_flow_spec_type type;
2017 struct ib_flow_spec_action_drop {
2018 enum ib_flow_spec_type type;
2022 struct ib_flow_spec_action_handle {
2023 enum ib_flow_spec_type type;
2025 struct ib_flow_action *act;
2028 enum ib_counters_description {
2033 struct ib_flow_spec_action_count {
2034 enum ib_flow_spec_type type;
2036 struct ib_counters *counters;
2039 union ib_flow_spec {
2044 struct ib_flow_spec_eth eth;
2045 struct ib_flow_spec_ib ib;
2046 struct ib_flow_spec_ipv4 ipv4;
2047 struct ib_flow_spec_tcp_udp tcp_udp;
2048 struct ib_flow_spec_ipv6 ipv6;
2049 struct ib_flow_spec_tunnel tunnel;
2050 struct ib_flow_spec_esp esp;
2051 struct ib_flow_spec_gre gre;
2052 struct ib_flow_spec_mpls mpls;
2053 struct ib_flow_spec_action_tag flow_tag;
2054 struct ib_flow_spec_action_drop drop;
2055 struct ib_flow_spec_action_handle action;
2056 struct ib_flow_spec_action_count flow_count;
2059 struct ib_flow_attr {
2060 enum ib_flow_attr_type type;
2066 union ib_flow_spec flows[];
2071 struct ib_device *device;
2072 struct ib_uobject *uobject;
2075 enum ib_flow_action_type {
2076 IB_FLOW_ACTION_UNSPECIFIED,
2077 IB_FLOW_ACTION_ESP = 1,
2080 struct ib_flow_action_attrs_esp_keymats {
2081 enum ib_uverbs_flow_action_esp_keymat protocol;
2083 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2087 struct ib_flow_action_attrs_esp_replays {
2088 enum ib_uverbs_flow_action_esp_replay protocol;
2090 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2094 enum ib_flow_action_attrs_esp_flags {
2095 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2096 * This is done in order to share the same flags between user-space and
2097 * kernel and spare an unnecessary translation.
2101 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2102 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2105 struct ib_flow_spec_list {
2106 struct ib_flow_spec_list *next;
2107 union ib_flow_spec spec;
2110 struct ib_flow_action_attrs_esp {
2111 struct ib_flow_action_attrs_esp_keymats *keymat;
2112 struct ib_flow_action_attrs_esp_replays *replay;
2113 struct ib_flow_spec_list *encap;
2114 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2115 * Value of 0 is a valid value.
2121 /* Use enum ib_flow_action_attrs_esp_flags */
2123 u64 hard_limit_pkts;
2126 struct ib_flow_action {
2127 struct ib_device *device;
2128 struct ib_uobject *uobject;
2129 enum ib_flow_action_type type;
2136 enum ib_process_mad_flags {
2137 IB_MAD_IGNORE_MKEY = 1,
2138 IB_MAD_IGNORE_BKEY = 2,
2139 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2142 enum ib_mad_result {
2143 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2144 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2145 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2146 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2149 struct ib_port_cache {
2151 struct ib_pkey_cache *pkey;
2152 struct ib_gid_table *gid;
2154 enum ib_port_state port_state;
2157 struct ib_port_immutable {
2164 struct ib_port_data {
2165 struct ib_device *ib_dev;
2167 struct ib_port_immutable immutable;
2169 spinlock_t pkey_list_lock;
2170 struct list_head pkey_list;
2172 struct ib_port_cache cache;
2174 spinlock_t netdev_lock;
2175 struct net_device __rcu *netdev;
2176 struct hlist_node ndev_hash_link;
2177 struct rdma_port_counter port_counter;
2178 struct rdma_hw_stats *hw_stats;
2181 /* rdma netdev type - specifies protocol type */
2182 enum rdma_netdev_t {
2183 RDMA_NETDEV_OPA_VNIC,
2188 * struct rdma_netdev - rdma netdev
2189 * For cases where netstack interfacing is required.
2191 struct rdma_netdev {
2193 struct ib_device *hca;
2198 * cleanup function must be specified.
2199 * FIXME: This is only used for OPA_VNIC and that usage should be
2202 void (*free_rdma_netdev)(struct net_device *netdev);
2204 /* control functions */
2205 void (*set_id)(struct net_device *netdev, int id);
2207 int (*send)(struct net_device *dev, struct sk_buff *skb,
2208 struct ib_ah *address, u32 dqpn);
2210 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2211 union ib_gid *gid, u16 mlid,
2212 int set_qkey, u32 qkey);
2213 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2214 union ib_gid *gid, u16 mlid);
2217 struct rdma_netdev_alloc_params {
2223 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2224 struct net_device *netdev, void *param);
2227 struct ib_odp_counters {
2229 atomic64_t invalidations;
2230 atomic64_t prefetch;
2233 struct ib_counters {
2234 struct ib_device *device;
2235 struct ib_uobject *uobject;
2236 /* num of objects attached */
2240 struct ib_counters_read_attr {
2243 u32 flags; /* use enum ib_read_counters_flags */
2246 struct uverbs_attr_bundle;
2248 struct iw_cm_conn_param;
2250 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2251 .size_##ib_struct = \
2252 (sizeof(struct drv_struct) + \
2253 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2254 BUILD_BUG_ON_ZERO( \
2255 !__same_type(((struct drv_struct *)NULL)->member, \
2258 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2259 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2261 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2262 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2264 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2266 struct rdma_user_mmap_entry {
2268 struct ib_ucontext *ucontext;
2269 unsigned long start_pgoff;
2271 bool driver_removed;
2274 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2276 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2278 return (u64)entry->start_pgoff << PAGE_SHIFT;
2282 * struct ib_device_ops - InfiniBand device operations
2283 * This structure defines all the InfiniBand device operations, providers will
2284 * need to define the supported operations, otherwise they will be set to null.
2286 struct ib_device_ops {
2287 struct module *owner;
2288 enum rdma_driver_id driver_id;
2290 unsigned int uverbs_no_driver_id_binding:1;
2292 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2293 const struct ib_send_wr **bad_send_wr);
2294 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2295 const struct ib_recv_wr **bad_recv_wr);
2296 void (*drain_rq)(struct ib_qp *qp);
2297 void (*drain_sq)(struct ib_qp *qp);
2298 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2299 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2300 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2301 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2302 int (*post_srq_recv)(struct ib_srq *srq,
2303 const struct ib_recv_wr *recv_wr,
2304 const struct ib_recv_wr **bad_recv_wr);
2305 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2306 u8 port_num, const struct ib_wc *in_wc,
2307 const struct ib_grh *in_grh,
2308 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2309 size_t *out_mad_size, u16 *out_mad_pkey_index);
2310 int (*query_device)(struct ib_device *device,
2311 struct ib_device_attr *device_attr,
2312 struct ib_udata *udata);
2313 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2314 struct ib_device_modify *device_modify);
2315 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2316 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2318 int (*query_port)(struct ib_device *device, u8 port_num,
2319 struct ib_port_attr *port_attr);
2320 int (*modify_port)(struct ib_device *device, u8 port_num,
2321 int port_modify_mask,
2322 struct ib_port_modify *port_modify);
2324 * The following mandatory functions are used only at device
2325 * registration. Keep functions such as these at the end of this
2326 * structure to avoid cache line misses when accessing struct ib_device
2329 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2330 struct ib_port_immutable *immutable);
2331 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2334 * When calling get_netdev, the HW vendor's driver should return the
2335 * net device of device @device at port @port_num or NULL if such
2336 * a net device doesn't exist. The vendor driver should call dev_hold
2337 * on this net device. The HW vendor's device driver must guarantee
2338 * that this function returns NULL before the net device has finished
2339 * NETDEV_UNREGISTER state.
2341 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2343 * rdma netdev operation
2345 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2346 * must return -EOPNOTSUPP if it doesn't support the specified type.
2348 struct net_device *(*alloc_rdma_netdev)(
2349 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2350 const char *name, unsigned char name_assign_type,
2351 void (*setup)(struct net_device *));
2353 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2354 enum rdma_netdev_t type,
2355 struct rdma_netdev_alloc_params *params);
2357 * query_gid should be return GID value for @device, when @port_num
2358 * link layer is either IB or iWarp. It is no-op if @port_num port
2359 * is RoCE link layer.
2361 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2364 * When calling add_gid, the HW vendor's driver should add the gid
2365 * of device of port at gid index available at @attr. Meta-info of
2366 * that gid (for example, the network device related to this gid) is
2367 * available at @attr. @context allows the HW vendor driver to store
2368 * extra information together with a GID entry. The HW vendor driver may
2369 * allocate memory to contain this information and store it in @context
2370 * when a new GID entry is written to. Params are consistent until the
2371 * next call of add_gid or delete_gid. The function should return 0 on
2372 * success or error otherwise. The function could be called
2373 * concurrently for different ports. This function is only called when
2374 * roce_gid_table is used.
2376 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2378 * When calling del_gid, the HW vendor's driver should delete the
2379 * gid of device @device at gid index gid_index of port port_num
2380 * available in @attr.
2381 * Upon the deletion of a GID entry, the HW vendor must free any
2382 * allocated memory. The caller will clear @context afterwards.
2383 * This function is only called when roce_gid_table is used.
2385 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2386 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2388 int (*alloc_ucontext)(struct ib_ucontext *context,
2389 struct ib_udata *udata);
2390 void (*dealloc_ucontext)(struct ib_ucontext *context);
2391 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2393 * This will be called once refcount of an entry in mmap_xa reaches
2394 * zero. The type of the memory that was mapped may differ between
2395 * entries and is opaque to the rdma_user_mmap interface.
2396 * Therefore needs to be implemented by the driver in mmap_free.
2398 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2399 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2400 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2401 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2402 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2403 struct ib_udata *udata);
2404 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2405 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2406 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2407 int (*create_srq)(struct ib_srq *srq,
2408 struct ib_srq_init_attr *srq_init_attr,
2409 struct ib_udata *udata);
2410 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2411 enum ib_srq_attr_mask srq_attr_mask,
2412 struct ib_udata *udata);
2413 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2414 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2415 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2416 struct ib_qp_init_attr *qp_init_attr,
2417 struct ib_udata *udata);
2418 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2419 int qp_attr_mask, struct ib_udata *udata);
2420 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2421 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2422 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2423 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2424 struct ib_udata *udata);
2425 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2426 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2427 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2428 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2429 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2430 u64 virt_addr, int mr_access_flags,
2431 struct ib_udata *udata);
2432 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2433 u64 virt_addr, int mr_access_flags,
2434 struct ib_pd *pd, struct ib_udata *udata);
2435 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2436 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2438 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2439 u32 max_num_data_sg,
2440 u32 max_num_meta_sg);
2441 int (*advise_mr)(struct ib_pd *pd,
2442 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2443 struct ib_sge *sg_list, u32 num_sge,
2444 struct uverbs_attr_bundle *attrs);
2445 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2446 unsigned int *sg_offset);
2447 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2448 struct ib_mr_status *mr_status);
2449 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2450 struct ib_udata *udata);
2451 int (*dealloc_mw)(struct ib_mw *mw);
2452 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2453 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2454 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2455 void (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2456 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2457 struct ib_flow_attr *flow_attr,
2458 struct ib_udata *udata);
2459 int (*destroy_flow)(struct ib_flow *flow_id);
2460 struct ib_flow_action *(*create_flow_action_esp)(
2461 struct ib_device *device,
2462 const struct ib_flow_action_attrs_esp *attr,
2463 struct uverbs_attr_bundle *attrs);
2464 int (*destroy_flow_action)(struct ib_flow_action *action);
2465 int (*modify_flow_action_esp)(
2466 struct ib_flow_action *action,
2467 const struct ib_flow_action_attrs_esp *attr,
2468 struct uverbs_attr_bundle *attrs);
2469 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2471 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2472 struct ifla_vf_info *ivf);
2473 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2474 struct ifla_vf_stats *stats);
2475 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2476 struct ifla_vf_guid *node_guid,
2477 struct ifla_vf_guid *port_guid);
2478 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2480 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2481 struct ib_wq_init_attr *init_attr,
2482 struct ib_udata *udata);
2483 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2484 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2485 u32 wq_attr_mask, struct ib_udata *udata);
2486 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2487 struct ib_device *device,
2488 struct ib_rwq_ind_table_init_attr *init_attr,
2489 struct ib_udata *udata);
2490 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2491 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2492 struct ib_ucontext *context,
2493 struct ib_dm_alloc_attr *attr,
2494 struct uverbs_attr_bundle *attrs);
2495 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2496 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2497 struct ib_dm_mr_attr *attr,
2498 struct uverbs_attr_bundle *attrs);
2499 int (*create_counters)(struct ib_counters *counters,
2500 struct uverbs_attr_bundle *attrs);
2501 void (*destroy_counters)(struct ib_counters *counters);
2502 int (*read_counters)(struct ib_counters *counters,
2503 struct ib_counters_read_attr *counters_read_attr,
2504 struct uverbs_attr_bundle *attrs);
2505 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2506 int data_sg_nents, unsigned int *data_sg_offset,
2507 struct scatterlist *meta_sg, int meta_sg_nents,
2508 unsigned int *meta_sg_offset);
2511 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2512 * driver initialized data. The struct is kfree()'ed by the sysfs
2513 * core when the device is removed. A lifespan of -1 in the return
2514 * struct tells the core to set a default lifespan.
2516 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2519 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2520 * @index - The index in the value array we wish to have updated, or
2521 * num_counters if we want all stats updated
2523 * < 0 - Error, no counters updated
2524 * index - Updated the single counter pointed to by index
2525 * num_counters - Updated all counters (will reset the timestamp
2526 * and prevent further calls for lifespan milliseconds)
2527 * Drivers are allowed to update all counters in leiu of just the
2528 * one given in index at their option
2530 int (*get_hw_stats)(struct ib_device *device,
2531 struct rdma_hw_stats *stats, u8 port, int index);
2533 * This function is called once for each port when a ib device is
2536 int (*init_port)(struct ib_device *device, u8 port_num,
2537 struct kobject *port_sysfs);
2539 * Allows rdma drivers to add their own restrack attributes.
2541 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2542 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2543 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2544 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2545 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2546 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2547 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2549 /* Device lifecycle callbacks */
2551 * Called after the device becomes registered, before clients are
2554 int (*enable_driver)(struct ib_device *dev);
2556 * This is called as part of ib_dealloc_device().
2558 void (*dealloc_driver)(struct ib_device *dev);
2560 /* iWarp CM callbacks */
2561 void (*iw_add_ref)(struct ib_qp *qp);
2562 void (*iw_rem_ref)(struct ib_qp *qp);
2563 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2564 int (*iw_connect)(struct iw_cm_id *cm_id,
2565 struct iw_cm_conn_param *conn_param);
2566 int (*iw_accept)(struct iw_cm_id *cm_id,
2567 struct iw_cm_conn_param *conn_param);
2568 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2570 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2571 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2573 * counter_bind_qp - Bind a QP to a counter.
2574 * @counter - The counter to be bound. If counter->id is zero then
2575 * the driver needs to allocate a new counter and set counter->id
2577 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2579 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2580 * counter and bind it onto the default one
2582 int (*counter_unbind_qp)(struct ib_qp *qp);
2584 * counter_dealloc -De-allocate the hw counter
2586 int (*counter_dealloc)(struct rdma_counter *counter);
2588 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2589 * the driver initialized data.
2591 struct rdma_hw_stats *(*counter_alloc_stats)(
2592 struct rdma_counter *counter);
2594 * counter_update_stats - Query the stats value of this counter
2596 int (*counter_update_stats)(struct rdma_counter *counter);
2599 * Allows rdma drivers to add their own restrack attributes
2600 * dumped via 'rdma stat' iproute2 command.
2602 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2604 /* query driver for its ucontext properties */
2605 int (*query_ucontext)(struct ib_ucontext *context,
2606 struct uverbs_attr_bundle *attrs);
2608 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2609 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2610 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2611 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2612 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2613 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2614 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2617 struct ib_core_device {
2618 /* device must be the first element in structure until,
2619 * union of ib_core_device and device exists in ib_device.
2622 possible_net_t rdma_net;
2623 struct kobject *ports_kobj;
2624 struct list_head port_list;
2625 struct ib_device *owner; /* reach back to owner ib_device */
2628 struct rdma_restrack_root;
2630 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2631 struct device *dma_device;
2632 struct ib_device_ops ops;
2633 char name[IB_DEVICE_NAME_MAX];
2634 struct rcu_head rcu_head;
2636 struct list_head event_handler_list;
2637 /* Protects event_handler_list */
2638 struct rw_semaphore event_handler_rwsem;
2640 /* Protects QP's event_handler calls and open_qp list */
2641 spinlock_t qp_open_list_lock;
2643 struct rw_semaphore client_data_rwsem;
2644 struct xarray client_data;
2645 struct mutex unregistration_lock;
2647 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2648 rwlock_t cache_lock;
2650 * port_data is indexed by port number
2652 struct ib_port_data *port_data;
2654 int num_comp_vectors;
2658 struct ib_core_device coredev;
2661 /* First group for device attributes,
2662 * Second group for driver provided attributes (optional).
2663 * It is NULL terminated array.
2665 const struct attribute_group *groups[3];
2667 u64 uverbs_cmd_mask;
2668 u64 uverbs_ex_cmd_mask;
2670 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2674 /* Indicates kernel verbs support, should not be used in drivers */
2675 u16 kverbs_provider:1;
2676 /* CQ adaptive moderation (RDMA DIM) */
2680 struct ib_device_attr attrs;
2681 struct attribute_group *hw_stats_ag;
2682 struct rdma_hw_stats *hw_stats;
2684 #ifdef CONFIG_CGROUP_RDMA
2685 struct rdmacg_device cg_device;
2690 spinlock_t cq_pools_lock;
2691 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2693 struct rdma_restrack_root *res;
2695 const struct uapi_definition *driver_def;
2698 * Positive refcount indicates that the device is currently
2699 * registered and cannot be unregistered.
2701 refcount_t refcount;
2702 struct completion unreg_completion;
2703 struct work_struct unregistration_work;
2705 const struct rdma_link_ops *link_ops;
2707 /* Protects compat_devs xarray modifications */
2708 struct mutex compat_devs_mutex;
2709 /* Maintains compat devices for each net namespace */
2710 struct xarray compat_devs;
2712 /* Used by iWarp CM */
2713 char iw_ifname[IFNAMSIZ];
2714 u32 iw_driver_flags;
2718 struct ib_client_nl_info;
2721 int (*add)(struct ib_device *ibdev);
2722 void (*remove)(struct ib_device *, void *client_data);
2723 void (*rename)(struct ib_device *dev, void *client_data);
2724 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2725 struct ib_client_nl_info *res);
2726 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2728 /* Returns the net_dev belonging to this ib_client and matching the
2730 * @dev: An RDMA device that the net_dev use for communication.
2731 * @port: A physical port number on the RDMA device.
2732 * @pkey: P_Key that the net_dev uses if applicable.
2733 * @gid: A GID that the net_dev uses to communicate.
2734 * @addr: An IP address the net_dev is configured with.
2735 * @client_data: The device's client data set by ib_set_client_data().
2737 * An ib_client that implements a net_dev on top of RDMA devices
2738 * (such as IP over IB) should implement this callback, allowing the
2739 * rdma_cm module to find the right net_dev for a given request.
2741 * The caller is responsible for calling dev_put on the returned
2743 struct net_device *(*get_net_dev_by_params)(
2744 struct ib_device *dev,
2747 const union ib_gid *gid,
2748 const struct sockaddr *addr,
2752 struct completion uses_zero;
2755 /* kverbs are not required by the client */
2760 * IB block DMA iterator
2762 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2763 * to a HW supported page size.
2765 struct ib_block_iter {
2766 /* internal states */
2767 struct scatterlist *__sg; /* sg holding the current aligned block */
2768 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2769 unsigned int __sg_nents; /* number of SG entries */
2770 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2771 unsigned int __pg_bit; /* alignment of current block */
2774 struct ib_device *_ib_alloc_device(size_t size);
2775 #define ib_alloc_device(drv_struct, member) \
2776 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2777 BUILD_BUG_ON_ZERO(offsetof( \
2778 struct drv_struct, member))), \
2779 struct drv_struct, member)
2781 void ib_dealloc_device(struct ib_device *device);
2783 void ib_get_device_fw_str(struct ib_device *device, char *str);
2785 int ib_register_device(struct ib_device *device, const char *name);
2786 void ib_unregister_device(struct ib_device *device);
2787 void ib_unregister_driver(enum rdma_driver_id driver_id);
2788 void ib_unregister_device_and_put(struct ib_device *device);
2789 void ib_unregister_device_queued(struct ib_device *ib_dev);
2791 int ib_register_client (struct ib_client *client);
2792 void ib_unregister_client(struct ib_client *client);
2794 void __rdma_block_iter_start(struct ib_block_iter *biter,
2795 struct scatterlist *sglist,
2797 unsigned long pgsz);
2798 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2801 * rdma_block_iter_dma_address - get the aligned dma address of the current
2802 * block held by the block iterator.
2803 * @biter: block iterator holding the memory block
2805 static inline dma_addr_t
2806 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2808 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2812 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2813 * @sglist: sglist to iterate over
2814 * @biter: block iterator holding the memory block
2815 * @nents: maximum number of sg entries to iterate over
2816 * @pgsz: best HW supported page size to use
2818 * Callers may use rdma_block_iter_dma_address() to get each
2819 * blocks aligned DMA address.
2821 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2822 for (__rdma_block_iter_start(biter, sglist, nents, \
2824 __rdma_block_iter_next(biter);)
2827 * ib_get_client_data - Get IB client context
2828 * @device:Device to get context for
2829 * @client:Client to get context for
2831 * ib_get_client_data() returns the client context data set with
2832 * ib_set_client_data(). This can only be called while the client is
2833 * registered to the device, once the ib_client remove() callback returns this
2836 static inline void *ib_get_client_data(struct ib_device *device,
2837 struct ib_client *client)
2839 return xa_load(&device->client_data, client->client_id);
2841 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2843 void ib_set_device_ops(struct ib_device *device,
2844 const struct ib_device_ops *ops);
2846 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2847 unsigned long pfn, unsigned long size, pgprot_t prot,
2848 struct rdma_user_mmap_entry *entry);
2849 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2850 struct rdma_user_mmap_entry *entry,
2852 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2853 struct rdma_user_mmap_entry *entry,
2854 size_t length, u32 min_pgoff,
2857 struct rdma_user_mmap_entry *
2858 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2859 unsigned long pgoff);
2860 struct rdma_user_mmap_entry *
2861 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2862 struct vm_area_struct *vma);
2863 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2865 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2867 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2869 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2872 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2874 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2877 static inline bool ib_is_buffer_cleared(const void __user *p,
2883 if (len > USHRT_MAX)
2886 buf = memdup_user(p, len);
2890 ret = !memchr_inv(buf, 0, len);
2895 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2899 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2903 * ib_is_destroy_retryable - Check whether the uobject destruction
2905 * @ret: The initial destruction return code
2906 * @why: remove reason
2907 * @uobj: The uobject that is destroyed
2909 * This function is a helper function that IB layer and low-level drivers
2910 * can use to consider whether the destruction of the given uobject is
2912 * It checks the original return code, if it wasn't success the destruction
2913 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2914 * the remove reason. (i.e. why).
2915 * Must be called with the object locked for destroy.
2917 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2918 struct ib_uobject *uobj)
2920 return ret && (why == RDMA_REMOVE_DESTROY ||
2921 uobj->context->cleanup_retryable);
2925 * ib_destroy_usecnt - Called during destruction to check the usecnt
2926 * @usecnt: The usecnt atomic
2927 * @why: remove reason
2928 * @uobj: The uobject that is destroyed
2930 * Non-zero usecnts will block destruction unless destruction was triggered by
2931 * a ucontext cleanup.
2933 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2934 enum rdma_remove_reason why,
2935 struct ib_uobject *uobj)
2937 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2943 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2944 * contains all required attributes and no attributes not allowed for
2945 * the given QP state transition.
2946 * @cur_state: Current QP state
2947 * @next_state: Next QP state
2949 * @mask: Mask of supplied QP attributes
2951 * This function is a helper function that a low-level driver's
2952 * modify_qp method can use to validate the consumer's input. It
2953 * checks that cur_state and next_state are valid QP states, that a
2954 * transition from cur_state to next_state is allowed by the IB spec,
2955 * and that the attribute mask supplied is allowed for the transition.
2957 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2958 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2960 void ib_register_event_handler(struct ib_event_handler *event_handler);
2961 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2962 void ib_dispatch_event(const struct ib_event *event);
2964 int ib_query_port(struct ib_device *device,
2965 u8 port_num, struct ib_port_attr *port_attr);
2967 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2971 * rdma_cap_ib_switch - Check if the device is IB switch
2972 * @device: Device to check
2974 * Device driver is responsible for setting is_switch bit on
2975 * in ib_device structure at init time.
2977 * Return: true if the device is IB switch.
2979 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2981 return device->is_switch;
2985 * rdma_start_port - Return the first valid port number for the device
2988 * @device: Device to be checked
2990 * Return start port number
2992 static inline u8 rdma_start_port(const struct ib_device *device)
2994 return rdma_cap_ib_switch(device) ? 0 : 1;
2998 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2999 * @device - The struct ib_device * to iterate over
3000 * @iter - The unsigned int to store the port number
3002 #define rdma_for_each_port(device, iter) \
3003 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3004 unsigned int, iter))); \
3005 iter <= rdma_end_port(device); (iter)++)
3008 * rdma_end_port - Return the last valid port number for the device
3011 * @device: Device to be checked
3013 * Return last port number
3015 static inline u8 rdma_end_port(const struct ib_device *device)
3017 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3020 static inline int rdma_is_port_valid(const struct ib_device *device,
3023 return (port >= rdma_start_port(device) &&
3024 port <= rdma_end_port(device));
3027 static inline bool rdma_is_grh_required(const struct ib_device *device,
3030 return device->port_data[port_num].immutable.core_cap_flags &
3031 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3034 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3036 return device->port_data[port_num].immutable.core_cap_flags &
3037 RDMA_CORE_CAP_PROT_IB;
3040 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3042 return device->port_data[port_num].immutable.core_cap_flags &
3043 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3046 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3048 return device->port_data[port_num].immutable.core_cap_flags &
3049 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3052 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3054 return device->port_data[port_num].immutable.core_cap_flags &
3055 RDMA_CORE_CAP_PROT_ROCE;
3058 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3060 return device->port_data[port_num].immutable.core_cap_flags &
3061 RDMA_CORE_CAP_PROT_IWARP;
3064 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3066 return rdma_protocol_ib(device, port_num) ||
3067 rdma_protocol_roce(device, port_num);
3070 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3072 return device->port_data[port_num].immutable.core_cap_flags &
3073 RDMA_CORE_CAP_PROT_RAW_PACKET;
3076 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3078 return device->port_data[port_num].immutable.core_cap_flags &
3079 RDMA_CORE_CAP_PROT_USNIC;
3083 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3084 * Management Datagrams.
3085 * @device: Device to check
3086 * @port_num: Port number to check
3088 * Management Datagrams (MAD) are a required part of the InfiniBand
3089 * specification and are supported on all InfiniBand devices. A slightly
3090 * extended version are also supported on OPA interfaces.
3092 * Return: true if the port supports sending/receiving of MAD packets.
3094 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3096 return device->port_data[port_num].immutable.core_cap_flags &
3097 RDMA_CORE_CAP_IB_MAD;
3101 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3102 * Management Datagrams.
3103 * @device: Device to check
3104 * @port_num: Port number to check
3106 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3107 * datagrams with their own versions. These OPA MADs share many but not all of
3108 * the characteristics of InfiniBand MADs.
3110 * OPA MADs differ in the following ways:
3112 * 1) MADs are variable size up to 2K
3113 * IBTA defined MADs remain fixed at 256 bytes
3114 * 2) OPA SMPs must carry valid PKeys
3115 * 3) OPA SMP packets are a different format
3117 * Return: true if the port supports OPA MAD packet formats.
3119 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3121 return device->port_data[port_num].immutable.core_cap_flags &
3122 RDMA_CORE_CAP_OPA_MAD;
3126 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3127 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3128 * @device: Device to check
3129 * @port_num: Port number to check
3131 * Each InfiniBand node is required to provide a Subnet Management Agent
3132 * that the subnet manager can access. Prior to the fabric being fully
3133 * configured by the subnet manager, the SMA is accessed via a well known
3134 * interface called the Subnet Management Interface (SMI). This interface
3135 * uses directed route packets to communicate with the SM to get around the
3136 * chicken and egg problem of the SM needing to know what's on the fabric
3137 * in order to configure the fabric, and needing to configure the fabric in
3138 * order to send packets to the devices on the fabric. These directed
3139 * route packets do not need the fabric fully configured in order to reach
3140 * their destination. The SMI is the only method allowed to send
3141 * directed route packets on an InfiniBand fabric.
3143 * Return: true if the port provides an SMI.
3145 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3147 return device->port_data[port_num].immutable.core_cap_flags &
3148 RDMA_CORE_CAP_IB_SMI;
3152 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3153 * Communication Manager.
3154 * @device: Device to check
3155 * @port_num: Port number to check
3157 * The InfiniBand Communication Manager is one of many pre-defined General
3158 * Service Agents (GSA) that are accessed via the General Service
3159 * Interface (GSI). It's role is to facilitate establishment of connections
3160 * between nodes as well as other management related tasks for established
3163 * Return: true if the port supports an IB CM (this does not guarantee that
3164 * a CM is actually running however).
3166 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3168 return device->port_data[port_num].immutable.core_cap_flags &
3169 RDMA_CORE_CAP_IB_CM;
3173 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3174 * Communication Manager.
3175 * @device: Device to check
3176 * @port_num: Port number to check
3178 * Similar to above, but specific to iWARP connections which have a different
3179 * managment protocol than InfiniBand.
3181 * Return: true if the port supports an iWARP CM (this does not guarantee that
3182 * a CM is actually running however).
3184 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3186 return device->port_data[port_num].immutable.core_cap_flags &
3187 RDMA_CORE_CAP_IW_CM;
3191 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3192 * Subnet Administration.
3193 * @device: Device to check
3194 * @port_num: Port number to check
3196 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3197 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3198 * fabrics, devices should resolve routes to other hosts by contacting the
3199 * SA to query the proper route.
3201 * Return: true if the port should act as a client to the fabric Subnet
3202 * Administration interface. This does not imply that the SA service is
3205 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3207 return device->port_data[port_num].immutable.core_cap_flags &
3208 RDMA_CORE_CAP_IB_SA;
3212 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3214 * @device: Device to check
3215 * @port_num: Port number to check
3217 * InfiniBand multicast registration is more complex than normal IPv4 or
3218 * IPv6 multicast registration. Each Host Channel Adapter must register
3219 * with the Subnet Manager when it wishes to join a multicast group. It
3220 * should do so only once regardless of how many queue pairs it subscribes
3221 * to this group. And it should leave the group only after all queue pairs
3222 * attached to the group have been detached.
3224 * Return: true if the port must undertake the additional adminstrative
3225 * overhead of registering/unregistering with the SM and tracking of the
3226 * total number of queue pairs attached to the multicast group.
3228 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3230 return rdma_cap_ib_sa(device, port_num);
3234 * rdma_cap_af_ib - Check if the port of device has the capability
3235 * Native Infiniband Address.
3236 * @device: Device to check
3237 * @port_num: Port number to check
3239 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3240 * GID. RoCE uses a different mechanism, but still generates a GID via
3241 * a prescribed mechanism and port specific data.
3243 * Return: true if the port uses a GID address to identify devices on the
3246 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3248 return device->port_data[port_num].immutable.core_cap_flags &
3249 RDMA_CORE_CAP_AF_IB;
3253 * rdma_cap_eth_ah - Check if the port of device has the capability
3254 * Ethernet Address Handle.
3255 * @device: Device to check
3256 * @port_num: Port number to check
3258 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3259 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3260 * port. Normally, packet headers are generated by the sending host
3261 * adapter, but when sending connectionless datagrams, we must manually
3262 * inject the proper headers for the fabric we are communicating over.
3264 * Return: true if we are running as a RoCE port and must force the
3265 * addition of a Global Route Header built from our Ethernet Address
3266 * Handle into our header list for connectionless packets.
3268 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3270 return device->port_data[port_num].immutable.core_cap_flags &
3271 RDMA_CORE_CAP_ETH_AH;
3275 * rdma_cap_opa_ah - Check if the port of device supports
3276 * OPA Address handles
3277 * @device: Device to check
3278 * @port_num: Port number to check
3280 * Return: true if we are running on an OPA device which supports
3281 * the extended OPA addressing.
3283 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3285 return (device->port_data[port_num].immutable.core_cap_flags &
3286 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3290 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3293 * @port_num: Port number
3295 * This MAD size includes the MAD headers and MAD payload. No other headers
3298 * Return the max MAD size required by the Port. Will return 0 if the port
3299 * does not support MADs
3301 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3303 return device->port_data[port_num].immutable.max_mad_size;
3307 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3308 * @device: Device to check
3309 * @port_num: Port number to check
3311 * RoCE GID table mechanism manages the various GIDs for a device.
3313 * NOTE: if allocating the port's GID table has failed, this call will still
3314 * return true, but any RoCE GID table API will fail.
3316 * Return: true if the port uses RoCE GID table mechanism in order to manage
3319 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3322 return rdma_protocol_roce(device, port_num) &&
3323 device->ops.add_gid && device->ops.del_gid;
3327 * Check if the device supports READ W/ INVALIDATE.
3329 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3332 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3333 * has support for it yet.
3335 return rdma_protocol_iwarp(dev, port_num);
3339 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3342 * @pgsz_bitmap: bitmap of HW supported page sizes
3344 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3345 unsigned long pgsz_bitmap)
3347 unsigned long align;
3350 align = addr & -addr;
3352 /* Find page bit such that addr is aligned to the highest supported
3355 pgsz = pgsz_bitmap & ~(-align << 1);
3357 return __ffs(pgsz_bitmap);
3363 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3365 * @port_num: 1 based Port number
3367 * Return true if port is an Intel OPA port , false if not
3369 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3372 return (device->port_data[port_num].immutable.core_cap_flags &
3373 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3377 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3379 * @port_num: Port number
3380 * @mtu: enum value of MTU
3382 * Return the MTU size supported by the port as an integer value. Will return
3383 * -1 if enum value of mtu is not supported.
3385 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3388 if (rdma_core_cap_opa_port(device, port))
3389 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3391 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3395 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3397 * @port_num: Port number
3398 * @attr: port attribute
3400 * Return the MTU size supported by the port as an integer value.
3402 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3403 struct ib_port_attr *attr)
3405 if (rdma_core_cap_opa_port(device, port))
3406 return attr->phys_mtu;
3408 return ib_mtu_enum_to_int(attr->max_mtu);
3411 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3413 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3414 struct ifla_vf_info *info);
3415 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3416 struct ifla_vf_stats *stats);
3417 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3418 struct ifla_vf_guid *node_guid,
3419 struct ifla_vf_guid *port_guid);
3420 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3423 int ib_query_pkey(struct ib_device *device,
3424 u8 port_num, u16 index, u16 *pkey);
3426 int ib_modify_device(struct ib_device *device,
3427 int device_modify_mask,
3428 struct ib_device_modify *device_modify);
3430 int ib_modify_port(struct ib_device *device,
3431 u8 port_num, int port_modify_mask,
3432 struct ib_port_modify *port_modify);
3434 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3435 u8 *port_num, u16 *index);
3437 int ib_find_pkey(struct ib_device *device,
3438 u8 port_num, u16 pkey, u16 *index);
3442 * Create a memory registration for all memory in the system and place
3443 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3444 * ULPs to avoid the overhead of dynamic MRs.
3446 * This flag is generally considered unsafe and must only be used in
3447 * extremly trusted environments. Every use of it will log a warning
3448 * in the kernel log.
3450 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3453 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3454 const char *caller);
3456 #define ib_alloc_pd(device, flags) \
3457 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3459 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3462 * ib_dealloc_pd - Deallocate kernel PD
3463 * @pd: The protection domain
3465 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3467 static inline void ib_dealloc_pd(struct ib_pd *pd)
3469 int ret = ib_dealloc_pd_user(pd, NULL);
3471 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3474 enum rdma_create_ah_flags {
3475 /* In a sleepable context */
3476 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3480 * rdma_create_ah - Creates an address handle for the given address vector.
3481 * @pd: The protection domain associated with the address handle.
3482 * @ah_attr: The attributes of the address vector.
3483 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3485 * The address handle is used to reference a local or global destination
3486 * in all UD QP post sends.
3488 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3492 * rdma_create_user_ah - Creates an address handle for the given address vector.
3493 * It resolves destination mac address for ah attribute of RoCE type.
3494 * @pd: The protection domain associated with the address handle.
3495 * @ah_attr: The attributes of the address vector.
3496 * @udata: pointer to user's input output buffer information need by
3499 * It returns 0 on success and returns appropriate error code on error.
3500 * The address handle is used to reference a local or global destination
3501 * in all UD QP post sends.
3503 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3504 struct rdma_ah_attr *ah_attr,
3505 struct ib_udata *udata);
3507 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3509 * @hdr: the L3 header to parse
3510 * @net_type: type of header to parse
3511 * @sgid: place to store source gid
3512 * @dgid: place to store destination gid
3514 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3515 enum rdma_network_type net_type,
3516 union ib_gid *sgid, union ib_gid *dgid);
3519 * ib_get_rdma_header_version - Get the header version
3520 * @hdr: the L3 header to parse
3522 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3525 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3527 * @device: Device on which the received message arrived.
3528 * @port_num: Port on which the received message arrived.
3529 * @wc: Work completion associated with the received message.
3530 * @grh: References the received global route header. This parameter is
3531 * ignored unless the work completion indicates that the GRH is valid.
3532 * @ah_attr: Returned attributes that can be used when creating an address
3533 * handle for replying to the message.
3534 * When ib_init_ah_attr_from_wc() returns success,
3535 * (a) for IB link layer it optionally contains a reference to SGID attribute
3536 * when GRH is present for IB link layer.
3537 * (b) for RoCE link layer it contains a reference to SGID attribute.
3538 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3539 * attributes which are initialized using ib_init_ah_attr_from_wc().
3542 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3543 const struct ib_wc *wc, const struct ib_grh *grh,
3544 struct rdma_ah_attr *ah_attr);
3547 * ib_create_ah_from_wc - Creates an address handle associated with the
3548 * sender of the specified work completion.
3549 * @pd: The protection domain associated with the address handle.
3550 * @wc: Work completion information associated with a received message.
3551 * @grh: References the received global route header. This parameter is
3552 * ignored unless the work completion indicates that the GRH is valid.
3553 * @port_num: The outbound port number to associate with the address.
3555 * The address handle is used to reference a local or global destination
3556 * in all UD QP post sends.
3558 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3559 const struct ib_grh *grh, u8 port_num);
3562 * rdma_modify_ah - Modifies the address vector associated with an address
3564 * @ah: The address handle to modify.
3565 * @ah_attr: The new address vector attributes to associate with the
3568 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3571 * rdma_query_ah - Queries the address vector associated with an address
3573 * @ah: The address handle to query.
3574 * @ah_attr: The address vector attributes associated with the address
3577 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3579 enum rdma_destroy_ah_flags {
3580 /* In a sleepable context */
3581 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3585 * rdma_destroy_ah_user - Destroys an address handle.
3586 * @ah: The address handle to destroy.
3587 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3588 * @udata: Valid user data or NULL for kernel objects
3590 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3593 * rdma_destroy_ah - Destroys an kernel address handle.
3594 * @ah: The address handle to destroy.
3595 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3597 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3599 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3601 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3603 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3606 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3607 struct ib_srq_init_attr *srq_init_attr,
3608 struct ib_usrq_object *uobject,
3609 struct ib_udata *udata);
3610 static inline struct ib_srq *
3611 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3613 if (!pd->device->ops.create_srq)
3614 return ERR_PTR(-EOPNOTSUPP);
3616 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3620 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3621 * @srq: The SRQ to modify.
3622 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3623 * the current values of selected SRQ attributes are returned.
3624 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3625 * are being modified.
3627 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3628 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3629 * the number of receives queued drops below the limit.
3631 int ib_modify_srq(struct ib_srq *srq,
3632 struct ib_srq_attr *srq_attr,
3633 enum ib_srq_attr_mask srq_attr_mask);
3636 * ib_query_srq - Returns the attribute list and current values for the
3638 * @srq: The SRQ to query.
3639 * @srq_attr: The attributes of the specified SRQ.
3641 int ib_query_srq(struct ib_srq *srq,
3642 struct ib_srq_attr *srq_attr);
3645 * ib_destroy_srq_user - Destroys the specified SRQ.
3646 * @srq: The SRQ to destroy.
3647 * @udata: Valid user data or NULL for kernel objects
3649 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3652 * ib_destroy_srq - Destroys the specified kernel SRQ.
3653 * @srq: The SRQ to destroy.
3655 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3657 static inline void ib_destroy_srq(struct ib_srq *srq)
3659 int ret = ib_destroy_srq_user(srq, NULL);
3661 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3665 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3666 * @srq: The SRQ to post the work request on.
3667 * @recv_wr: A list of work requests to post on the receive queue.
3668 * @bad_recv_wr: On an immediate failure, this parameter will reference
3669 * the work request that failed to be posted on the QP.
3671 static inline int ib_post_srq_recv(struct ib_srq *srq,
3672 const struct ib_recv_wr *recv_wr,
3673 const struct ib_recv_wr **bad_recv_wr)
3675 const struct ib_recv_wr *dummy;
3677 return srq->device->ops.post_srq_recv(srq, recv_wr,
3678 bad_recv_wr ? : &dummy);
3681 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3682 struct ib_qp_init_attr *qp_init_attr);
3685 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3686 * @qp: The QP to modify.
3687 * @attr: On input, specifies the QP attributes to modify. On output,
3688 * the current values of selected QP attributes are returned.
3689 * @attr_mask: A bit-mask used to specify which attributes of the QP
3690 * are being modified.
3691 * @udata: pointer to user's input output buffer information
3692 * are being modified.
3693 * It returns 0 on success and returns appropriate error code on error.
3695 int ib_modify_qp_with_udata(struct ib_qp *qp,
3696 struct ib_qp_attr *attr,
3698 struct ib_udata *udata);
3701 * ib_modify_qp - Modifies the attributes for the specified QP and then
3702 * transitions the QP to the given state.
3703 * @qp: The QP to modify.
3704 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3705 * the current values of selected QP attributes are returned.
3706 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3707 * are being modified.
3709 int ib_modify_qp(struct ib_qp *qp,
3710 struct ib_qp_attr *qp_attr,
3714 * ib_query_qp - Returns the attribute list and current values for the
3716 * @qp: The QP to query.
3717 * @qp_attr: The attributes of the specified QP.
3718 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3719 * @qp_init_attr: Additional attributes of the selected QP.
3721 * The qp_attr_mask may be used to limit the query to gathering only the
3722 * selected attributes.
3724 int ib_query_qp(struct ib_qp *qp,
3725 struct ib_qp_attr *qp_attr,
3727 struct ib_qp_init_attr *qp_init_attr);
3730 * ib_destroy_qp - Destroys the specified QP.
3731 * @qp: The QP to destroy.
3732 * @udata: Valid udata or NULL for kernel objects
3734 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3737 * ib_destroy_qp - Destroys the specified kernel QP.
3738 * @qp: The QP to destroy.
3740 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3742 static inline int ib_destroy_qp(struct ib_qp *qp)
3744 return ib_destroy_qp_user(qp, NULL);
3748 * ib_open_qp - Obtain a reference to an existing sharable QP.
3749 * @xrcd - XRC domain
3750 * @qp_open_attr: Attributes identifying the QP to open.
3752 * Returns a reference to a sharable QP.
3754 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3755 struct ib_qp_open_attr *qp_open_attr);
3758 * ib_close_qp - Release an external reference to a QP.
3759 * @qp: The QP handle to release
3761 * The opened QP handle is released by the caller. The underlying
3762 * shared QP is not destroyed until all internal references are released.
3764 int ib_close_qp(struct ib_qp *qp);
3767 * ib_post_send - Posts a list of work requests to the send queue of
3769 * @qp: The QP to post the work request on.
3770 * @send_wr: A list of work requests to post on the send queue.
3771 * @bad_send_wr: On an immediate failure, this parameter will reference
3772 * the work request that failed to be posted on the QP.
3774 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3775 * error is returned, the QP state shall not be affected,
3776 * ib_post_send() will return an immediate error after queueing any
3777 * earlier work requests in the list.
3779 static inline int ib_post_send(struct ib_qp *qp,
3780 const struct ib_send_wr *send_wr,
3781 const struct ib_send_wr **bad_send_wr)
3783 const struct ib_send_wr *dummy;
3785 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3789 * ib_post_recv - Posts a list of work requests to the receive queue of
3791 * @qp: The QP to post the work request on.
3792 * @recv_wr: A list of work requests to post on the receive queue.
3793 * @bad_recv_wr: On an immediate failure, this parameter will reference
3794 * the work request that failed to be posted on the QP.
3796 static inline int ib_post_recv(struct ib_qp *qp,
3797 const struct ib_recv_wr *recv_wr,
3798 const struct ib_recv_wr **bad_recv_wr)
3800 const struct ib_recv_wr *dummy;
3802 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3805 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3806 int comp_vector, enum ib_poll_context poll_ctx,
3807 const char *caller);
3808 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3809 int nr_cqe, int comp_vector,
3810 enum ib_poll_context poll_ctx)
3812 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3816 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3817 int nr_cqe, enum ib_poll_context poll_ctx,
3818 const char *caller);
3821 * ib_alloc_cq_any: Allocate kernel CQ
3822 * @dev: The IB device
3823 * @private: Private data attached to the CQE
3824 * @nr_cqe: Number of CQEs in the CQ
3825 * @poll_ctx: Context used for polling the CQ
3827 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3828 void *private, int nr_cqe,
3829 enum ib_poll_context poll_ctx)
3831 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3835 void ib_free_cq(struct ib_cq *cq);
3836 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3839 * ib_create_cq - Creates a CQ on the specified device.
3840 * @device: The device on which to create the CQ.
3841 * @comp_handler: A user-specified callback that is invoked when a
3842 * completion event occurs on the CQ.
3843 * @event_handler: A user-specified callback that is invoked when an
3844 * asynchronous event not associated with a completion occurs on the CQ.
3845 * @cq_context: Context associated with the CQ returned to the user via
3846 * the associated completion and event handlers.
3847 * @cq_attr: The attributes the CQ should be created upon.
3849 * Users can examine the cq structure to determine the actual CQ size.
3851 struct ib_cq *__ib_create_cq(struct ib_device *device,
3852 ib_comp_handler comp_handler,
3853 void (*event_handler)(struct ib_event *, void *),
3855 const struct ib_cq_init_attr *cq_attr,
3856 const char *caller);
3857 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3858 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3861 * ib_resize_cq - Modifies the capacity of the CQ.
3862 * @cq: The CQ to resize.
3863 * @cqe: The minimum size of the CQ.
3865 * Users can examine the cq structure to determine the actual CQ size.
3867 int ib_resize_cq(struct ib_cq *cq, int cqe);
3870 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3871 * @cq: The CQ to modify.
3872 * @cq_count: number of CQEs that will trigger an event
3873 * @cq_period: max period of time in usec before triggering an event
3876 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3879 * ib_destroy_cq_user - Destroys the specified CQ.
3880 * @cq: The CQ to destroy.
3881 * @udata: Valid user data or NULL for kernel objects
3883 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3886 * ib_destroy_cq - Destroys the specified kernel CQ.
3887 * @cq: The CQ to destroy.
3889 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3891 static inline void ib_destroy_cq(struct ib_cq *cq)
3893 ib_destroy_cq_user(cq, NULL);
3897 * ib_poll_cq - poll a CQ for completion(s)
3898 * @cq:the CQ being polled
3899 * @num_entries:maximum number of completions to return
3900 * @wc:array of at least @num_entries &struct ib_wc where completions
3903 * Poll a CQ for (possibly multiple) completions. If the return value
3904 * is < 0, an error occurred. If the return value is >= 0, it is the
3905 * number of completions returned. If the return value is
3906 * non-negative and < num_entries, then the CQ was emptied.
3908 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3911 return cq->device->ops.poll_cq(cq, num_entries, wc);
3915 * ib_req_notify_cq - Request completion notification on a CQ.
3916 * @cq: The CQ to generate an event for.
3918 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3919 * to request an event on the next solicited event or next work
3920 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3921 * may also be |ed in to request a hint about missed events, as
3925 * < 0 means an error occurred while requesting notification
3926 * == 0 means notification was requested successfully, and if
3927 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3928 * were missed and it is safe to wait for another event. In
3929 * this case is it guaranteed that any work completions added
3930 * to the CQ since the last CQ poll will trigger a completion
3931 * notification event.
3932 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3933 * in. It means that the consumer must poll the CQ again to
3934 * make sure it is empty to avoid missing an event because of a
3935 * race between requesting notification and an entry being
3936 * added to the CQ. This return value means it is possible
3937 * (but not guaranteed) that a work completion has been added
3938 * to the CQ since the last poll without triggering a
3939 * completion notification event.
3941 static inline int ib_req_notify_cq(struct ib_cq *cq,
3942 enum ib_cq_notify_flags flags)
3944 return cq->device->ops.req_notify_cq(cq, flags);
3947 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3948 int comp_vector_hint,
3949 enum ib_poll_context poll_ctx);
3951 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3954 * ib_req_ncomp_notif - Request completion notification when there are
3955 * at least the specified number of unreaped completions on the CQ.
3956 * @cq: The CQ to generate an event for.
3957 * @wc_cnt: The number of unreaped completions that should be on the
3958 * CQ before an event is generated.
3960 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3962 return cq->device->ops.req_ncomp_notif ?
3963 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3968 * ib_dma_mapping_error - check a DMA addr for error
3969 * @dev: The device for which the dma_addr was created
3970 * @dma_addr: The DMA address to check
3972 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3974 return dma_mapping_error(dev->dma_device, dma_addr);
3978 * ib_dma_map_single - Map a kernel virtual address to DMA address
3979 * @dev: The device for which the dma_addr is to be created
3980 * @cpu_addr: The kernel virtual address
3981 * @size: The size of the region in bytes
3982 * @direction: The direction of the DMA
3984 static inline u64 ib_dma_map_single(struct ib_device *dev,
3985 void *cpu_addr, size_t size,
3986 enum dma_data_direction direction)
3988 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3992 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3993 * @dev: The device for which the DMA address was created
3994 * @addr: The DMA address
3995 * @size: The size of the region in bytes
3996 * @direction: The direction of the DMA
3998 static inline void ib_dma_unmap_single(struct ib_device *dev,
3999 u64 addr, size_t size,
4000 enum dma_data_direction direction)
4002 dma_unmap_single(dev->dma_device, addr, size, direction);
4006 * ib_dma_map_page - Map a physical page to DMA address
4007 * @dev: The device for which the dma_addr is to be created
4008 * @page: The page to be mapped
4009 * @offset: The offset within the page
4010 * @size: The size of the region in bytes
4011 * @direction: The direction of the DMA
4013 static inline u64 ib_dma_map_page(struct ib_device *dev,
4015 unsigned long offset,
4017 enum dma_data_direction direction)
4019 return dma_map_page(dev->dma_device, page, offset, size, direction);
4023 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4024 * @dev: The device for which the DMA address was created
4025 * @addr: The DMA address
4026 * @size: The size of the region in bytes
4027 * @direction: The direction of the DMA
4029 static inline void ib_dma_unmap_page(struct ib_device *dev,
4030 u64 addr, size_t size,
4031 enum dma_data_direction direction)
4033 dma_unmap_page(dev->dma_device, addr, size, direction);
4037 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4038 * @dev: The device for which the DMA addresses are to be created
4039 * @sg: The array of scatter/gather entries
4040 * @nents: The number of scatter/gather entries
4041 * @direction: The direction of the DMA
4043 static inline int ib_dma_map_sg(struct ib_device *dev,
4044 struct scatterlist *sg, int nents,
4045 enum dma_data_direction direction)
4047 return dma_map_sg(dev->dma_device, sg, nents, direction);
4051 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4052 * @dev: The device for which the DMA addresses were created
4053 * @sg: The array of scatter/gather entries
4054 * @nents: The number of scatter/gather entries
4055 * @direction: The direction of the DMA
4057 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4058 struct scatterlist *sg, int nents,
4059 enum dma_data_direction direction)
4061 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4064 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4065 struct scatterlist *sg, int nents,
4066 enum dma_data_direction direction,
4067 unsigned long dma_attrs)
4069 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4073 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4074 struct scatterlist *sg, int nents,
4075 enum dma_data_direction direction,
4076 unsigned long dma_attrs)
4078 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4082 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4083 * @dev: The device to query
4085 * The returned value represents a size in bytes.
4087 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4089 return dma_get_max_seg_size(dev->dma_device);
4093 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4094 * @dev: The device for which the DMA address was created
4095 * @addr: The DMA address
4096 * @size: The size of the region in bytes
4097 * @dir: The direction of the DMA
4099 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4102 enum dma_data_direction dir)
4104 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4108 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4109 * @dev: The device for which the DMA address was created
4110 * @addr: The DMA address
4111 * @size: The size of the region in bytes
4112 * @dir: The direction of the DMA
4114 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4117 enum dma_data_direction dir)
4119 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4123 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4124 * @dev: The device for which the DMA address is requested
4125 * @size: The size of the region to allocate in bytes
4126 * @dma_handle: A pointer for returning the DMA address of the region
4127 * @flag: memory allocator flags
4129 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4131 dma_addr_t *dma_handle,
4134 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4138 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4139 * @dev: The device for which the DMA addresses were allocated
4140 * @size: The size of the region
4141 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4142 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4144 static inline void ib_dma_free_coherent(struct ib_device *dev,
4145 size_t size, void *cpu_addr,
4146 dma_addr_t dma_handle)
4148 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4151 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4152 * space. This function should be called when 'current' is the owning MM.
4154 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4155 u64 virt_addr, int mr_access_flags);
4157 /* ib_advise_mr - give an advice about an address range in a memory region */
4158 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4159 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4161 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4162 * HCA translation table.
4163 * @mr: The memory region to deregister.
4164 * @udata: Valid user data or NULL for kernel object
4166 * This function can fail, if the memory region has memory windows bound to it.
4168 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4171 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4172 * HCA translation table.
4173 * @mr: The memory region to deregister.
4175 * This function can fail, if the memory region has memory windows bound to it.
4177 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4179 static inline int ib_dereg_mr(struct ib_mr *mr)
4181 return ib_dereg_mr_user(mr, NULL);
4184 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4187 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4188 u32 max_num_data_sg,
4189 u32 max_num_meta_sg);
4192 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4194 * @mr - struct ib_mr pointer to be updated.
4195 * @newkey - new key to be used.
4197 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4199 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4200 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4204 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4205 * for calculating a new rkey for type 2 memory windows.
4206 * @rkey - the rkey to increment.
4208 static inline u32 ib_inc_rkey(u32 rkey)
4210 const u32 mask = 0x000000ff;
4211 return ((rkey + 1) & mask) | (rkey & ~mask);
4215 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4216 * @qp: QP to attach to the multicast group. The QP must be type
4218 * @gid: Multicast group GID.
4219 * @lid: Multicast group LID in host byte order.
4221 * In order to send and receive multicast packets, subnet
4222 * administration must have created the multicast group and configured
4223 * the fabric appropriately. The port associated with the specified
4224 * QP must also be a member of the multicast group.
4226 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4229 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4230 * @qp: QP to detach from the multicast group.
4231 * @gid: Multicast group GID.
4232 * @lid: Multicast group LID in host byte order.
4234 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4236 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4237 struct inode *inode, struct ib_udata *udata);
4238 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4240 static inline int ib_check_mr_access(int flags)
4243 * Local write permission is required if remote write or
4244 * remote atomic permission is also requested.
4246 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4247 !(flags & IB_ACCESS_LOCAL_WRITE))
4250 if (flags & ~IB_ACCESS_SUPPORTED)
4256 static inline bool ib_access_writable(int access_flags)
4259 * We have writable memory backing the MR if any of the following
4260 * access flags are set. "Local write" and "remote write" obviously
4261 * require write access. "Remote atomic" can do things like fetch and
4262 * add, which will modify memory, and "MW bind" can change permissions
4263 * by binding a window.
4265 return access_flags &
4266 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4267 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4271 * ib_check_mr_status: lightweight check of MR status.
4272 * This routine may provide status checks on a selected
4273 * ib_mr. first use is for signature status check.
4275 * @mr: A memory region.
4276 * @check_mask: Bitmask of which checks to perform from
4277 * ib_mr_status_check enumeration.
4278 * @mr_status: The container of relevant status checks.
4279 * failed checks will be indicated in the status bitmask
4280 * and the relevant info shall be in the error item.
4282 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4283 struct ib_mr_status *mr_status);
4286 * ib_device_try_get: Hold a registration lock
4287 * device: The device to lock
4289 * A device under an active registration lock cannot become unregistered. It
4290 * is only possible to obtain a registration lock on a device that is fully
4291 * registered, otherwise this function returns false.
4293 * The registration lock is only necessary for actions which require the
4294 * device to still be registered. Uses that only require the device pointer to
4295 * be valid should use get_device(&ibdev->dev) to hold the memory.
4298 static inline bool ib_device_try_get(struct ib_device *dev)
4300 return refcount_inc_not_zero(&dev->refcount);
4303 void ib_device_put(struct ib_device *device);
4304 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4305 enum rdma_driver_id driver_id);
4306 struct ib_device *ib_device_get_by_name(const char *name,
4307 enum rdma_driver_id driver_id);
4308 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4309 u16 pkey, const union ib_gid *gid,
4310 const struct sockaddr *addr);
4311 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4313 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4315 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4316 struct ib_wq_init_attr *init_attr);
4317 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4318 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4320 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4322 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4323 unsigned int *sg_offset, unsigned int page_size);
4324 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4325 int data_sg_nents, unsigned int *data_sg_offset,
4326 struct scatterlist *meta_sg, int meta_sg_nents,
4327 unsigned int *meta_sg_offset, unsigned int page_size);
4330 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4331 unsigned int *sg_offset, unsigned int page_size)
4335 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4341 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4342 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4344 void ib_drain_rq(struct ib_qp *qp);
4345 void ib_drain_sq(struct ib_qp *qp);
4346 void ib_drain_qp(struct ib_qp *qp);
4348 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4350 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4352 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4353 return attr->roce.dmac;
4357 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4359 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4360 attr->ib.dlid = (u16)dlid;
4361 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4362 attr->opa.dlid = dlid;
4365 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4367 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4368 return attr->ib.dlid;
4369 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4370 return attr->opa.dlid;
4374 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4379 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4384 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4387 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4388 attr->ib.src_path_bits = src_path_bits;
4389 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4390 attr->opa.src_path_bits = src_path_bits;
4393 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4395 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4396 return attr->ib.src_path_bits;
4397 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4398 return attr->opa.src_path_bits;
4402 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4405 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4406 attr->opa.make_grd = make_grd;
4409 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4411 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4412 return attr->opa.make_grd;
4416 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4418 attr->port_num = port_num;
4421 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4423 return attr->port_num;
4426 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4429 attr->static_rate = static_rate;
4432 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4434 return attr->static_rate;
4437 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4438 enum ib_ah_flags flag)
4440 attr->ah_flags = flag;
4443 static inline enum ib_ah_flags
4444 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4446 return attr->ah_flags;
4449 static inline const struct ib_global_route
4450 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4455 /*To retrieve and modify the grh */
4456 static inline struct ib_global_route
4457 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4462 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4464 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4466 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4469 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4472 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4474 grh->dgid.global.subnet_prefix = prefix;
4477 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4480 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4482 grh->dgid.global.interface_id = if_id;
4485 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4486 union ib_gid *dgid, u32 flow_label,
4487 u8 sgid_index, u8 hop_limit,
4490 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4492 attr->ah_flags = IB_AH_GRH;
4495 grh->flow_label = flow_label;
4496 grh->sgid_index = sgid_index;
4497 grh->hop_limit = hop_limit;
4498 grh->traffic_class = traffic_class;
4499 grh->sgid_attr = NULL;
4502 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4503 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4504 u32 flow_label, u8 hop_limit, u8 traffic_class,
4505 const struct ib_gid_attr *sgid_attr);
4506 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4507 const struct rdma_ah_attr *src);
4508 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4509 const struct rdma_ah_attr *new);
4510 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4513 * rdma_ah_find_type - Return address handle type.
4515 * @dev: Device to be checked
4516 * @port_num: Port number
4518 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4521 if (rdma_protocol_roce(dev, port_num))
4522 return RDMA_AH_ATTR_TYPE_ROCE;
4523 if (rdma_protocol_ib(dev, port_num)) {
4524 if (rdma_cap_opa_ah(dev, port_num))
4525 return RDMA_AH_ATTR_TYPE_OPA;
4526 return RDMA_AH_ATTR_TYPE_IB;
4529 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4533 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4534 * In the current implementation the only way to get
4535 * get the 32bit lid is from other sources for OPA.
4536 * For IB, lids will always be 16bits so cast the
4537 * value accordingly.
4541 static inline u16 ib_lid_cpu16(u32 lid)
4543 WARN_ON_ONCE(lid & 0xFFFF0000);
4548 * ib_lid_be16 - Return lid in 16bit BE encoding.
4552 static inline __be16 ib_lid_be16(u32 lid)
4554 WARN_ON_ONCE(lid & 0xFFFF0000);
4555 return cpu_to_be16((u16)lid);
4559 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4561 * @device: the rdma device
4562 * @comp_vector: index of completion vector
4564 * Returns NULL on failure, otherwise a corresponding cpu map of the
4565 * completion vector (returns all-cpus map if the device driver doesn't
4566 * implement get_vector_affinity).
4568 static inline const struct cpumask *
4569 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4571 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4572 !device->ops.get_vector_affinity)
4575 return device->ops.get_vector_affinity(device, comp_vector);
4580 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4581 * and add their gids, as needed, to the relevant RoCE devices.
4583 * @device: the rdma device
4585 void rdma_roce_rescan_device(struct ib_device *ibdev);
4587 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4589 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4591 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4592 enum rdma_netdev_t type, const char *name,
4593 unsigned char name_assign_type,
4594 void (*setup)(struct net_device *));
4596 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4597 enum rdma_netdev_t type, const char *name,
4598 unsigned char name_assign_type,
4599 void (*setup)(struct net_device *),
4600 struct net_device *netdev);
4603 * rdma_set_device_sysfs_group - Set device attributes group to have
4604 * driver specific sysfs entries at
4605 * for infiniband class.
4607 * @device: device pointer for which attributes to be created
4608 * @group: Pointer to group which should be added when device
4609 * is registered with sysfs.
4610 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4611 * group per device to have sysfs attributes.
4613 * NOTE: New drivers should not make use of this API; instead new device
4614 * parameter should be exposed via netlink command. This API and mechanism
4615 * exist only for existing drivers.
4618 rdma_set_device_sysfs_group(struct ib_device *dev,
4619 const struct attribute_group *group)
4621 dev->groups[1] = group;
4625 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4627 * @device: device pointer for which ib_device pointer to retrieve
4629 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4632 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4634 struct ib_core_device *coredev =
4635 container_of(device, struct ib_core_device, dev);
4637 return coredev->owner;
4641 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4642 * ib_device holder structure from device pointer.
4644 * NOTE: New drivers should not make use of this API; This API is only for
4645 * existing drivers who have exposed sysfs entries using
4646 * rdma_set_device_sysfs_group().
4648 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4649 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4651 bool rdma_dev_access_netns(const struct ib_device *device,
4652 const struct net *net);
4654 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4655 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4656 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4659 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4662 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4663 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4666 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4668 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4670 fl_low ^= fl_high >> 14;
4671 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4675 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4676 * local and remote qpn values
4678 * This function folded the multiplication results of two qpns, 24 bit each,
4679 * fields, and converts it to a 20 bit results.
4681 * This function will create symmetric flow_label value based on the local
4682 * and remote qpn values. this will allow both the requester and responder
4683 * to calculate the same flow_label for a given connection.
4685 * This helper function should be used by driver in case the upper layer
4686 * provide a zero flow_label value. This is to improve entropy of RDMA
4687 * traffic in the network.
4689 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4691 u64 v = (u64)lqpn * rqpn;
4696 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4698 #endif /* IB_VERBS_H */