1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
3 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
5 * Copyright (c) 2004, 2020 Intel Corporation. All rights reserved.
6 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
7 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
8 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
46 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
50 struct ib_usrq_object;
54 struct hw_stats_device_data;
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 const char *format, ...);
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...) \
83 dynamic_ibdev_dbg(__dev, format, ##args)
87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
92 static DEFINE_RATELIMIT_STATE(_rs, \
93 DEFAULT_RATELIMIT_INTERVAL, \
94 DEFAULT_RATELIMIT_BURST); \
95 if (__ratelimit(&_rs)) \
96 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
119 static DEFINE_RATELIMIT_STATE(_rs, \
120 DEFAULT_RATELIMIT_INTERVAL, \
121 DEFAULT_RATELIMIT_BURST); \
122 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
123 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
124 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
128 __printf(2, 3) __cold
130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
136 __be64 subnet_prefix;
141 extern union ib_gid zgid;
144 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
150 #define ROCE_V2_UDP_DPORT 4791
152 struct net_device __rcu *ndev;
153 struct ib_device *device;
155 enum ib_gid_type gid_type;
161 /* set the local administered indication */
162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
165 enum rdma_transport_type {
167 RDMA_TRANSPORT_IWARP,
168 RDMA_TRANSPORT_USNIC,
169 RDMA_TRANSPORT_USNIC_UDP,
170 RDMA_TRANSPORT_UNSPECIFIED,
173 enum rdma_protocol_type {
177 RDMA_PROTOCOL_USNIC_UDP
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
183 enum rdma_network_type {
185 RDMA_NETWORK_ROCE_V1,
190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
192 if (network_type == RDMA_NETWORK_IPV4 ||
193 network_type == RDMA_NETWORK_IPV6)
194 return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 else if (network_type == RDMA_NETWORK_ROCE_V1)
196 return IB_GID_TYPE_ROCE;
198 return IB_GID_TYPE_IB;
201 static inline enum rdma_network_type
202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
204 if (attr->gid_type == IB_GID_TYPE_IB)
205 return RDMA_NETWORK_IB;
207 if (attr->gid_type == IB_GID_TYPE_ROCE)
208 return RDMA_NETWORK_ROCE_V1;
210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 return RDMA_NETWORK_IPV4;
213 return RDMA_NETWORK_IPV6;
216 enum rdma_link_layer {
217 IB_LINK_LAYER_UNSPECIFIED,
218 IB_LINK_LAYER_INFINIBAND,
219 IB_LINK_LAYER_ETHERNET,
222 enum ib_device_cap_flags {
223 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
224 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
225 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
226 IB_DEVICE_RAW_MULTI = (1 << 3),
227 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
228 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
229 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
230 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
231 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
232 /* Not in use, former INIT_TYPE = (1 << 9),*/
233 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
234 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
235 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
236 IB_DEVICE_SRQ_RESIZE = (1 << 13),
237 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
240 * This device supports a per-device lkey or stag that can be
241 * used without performing a memory registration for the local
242 * memory. Note that ULPs should never check this flag, but
243 * instead of use the local_dma_lkey flag in the ib_pd structure,
244 * which will always contain a usable lkey.
246 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
247 /* Reserved, old SEND_W_INV = (1 << 16),*/
248 IB_DEVICE_MEM_WINDOW = (1 << 17),
250 * Devices should set IB_DEVICE_UD_IP_SUM if they support
251 * insertion of UDP and TCP checksum on outgoing UD IPoIB
252 * messages and can verify the validity of checksum for
253 * incoming messages. Setting this flag implies that the
254 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
256 IB_DEVICE_UD_IP_CSUM = (1 << 18),
257 IB_DEVICE_UD_TSO = (1 << 19),
258 IB_DEVICE_XRC = (1 << 20),
261 * This device supports the IB "base memory management extension",
262 * which includes support for fast registrations (IB_WR_REG_MR,
263 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
264 * also be set by any iWarp device which must support FRs to comply
265 * to the iWarp verbs spec. iWarp devices also support the
266 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
269 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
270 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
271 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
272 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
273 IB_DEVICE_RC_IP_CSUM = (1 << 25),
274 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
275 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
277 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
278 * support execution of WQEs that involve synchronization
279 * of I/O operations with single completion queue managed
282 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
283 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
284 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
285 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
286 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
287 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
288 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
289 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
290 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
291 /* The device supports padding incoming writes to cacheline. */
292 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
293 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
302 enum ib_odp_general_cap_bits {
303 IB_ODP_SUPPORT = 1 << 0,
304 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
307 enum ib_odp_transport_cap_bits {
308 IB_ODP_SUPPORT_SEND = 1 << 0,
309 IB_ODP_SUPPORT_RECV = 1 << 1,
310 IB_ODP_SUPPORT_WRITE = 1 << 2,
311 IB_ODP_SUPPORT_READ = 1 << 3,
312 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
313 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
317 uint64_t general_caps;
319 uint32_t rc_odp_caps;
320 uint32_t uc_odp_caps;
321 uint32_t ud_odp_caps;
322 uint32_t xrc_odp_caps;
323 } per_transport_caps;
327 /* Corresponding bit will be set if qp type from
328 * 'enum ib_qp_type' is supported, e.g.
329 * supported_qpts |= 1 << IB_QPT_UD
332 u32 max_rwq_indirection_tables;
333 u32 max_rwq_indirection_table_size;
336 enum ib_tm_cap_flags {
337 /* Support tag matching with rendezvous offload for RC transport */
338 IB_TM_CAP_RNDV_RC = 1 << 0,
342 /* Max size of RNDV header */
343 u32 max_rndv_hdr_size;
344 /* Max number of entries in tag matching list */
346 /* From enum ib_tm_cap_flags */
348 /* Max number of outstanding list operations */
350 /* Max number of SGE in tag matching entry */
354 struct ib_cq_init_attr {
360 enum ib_cq_attr_mask {
361 IB_CQ_MODERATE = 1 << 0,
365 u16 max_cq_moderation_count;
366 u16 max_cq_moderation_period;
369 struct ib_dm_mr_attr {
375 struct ib_dm_alloc_attr {
381 struct ib_device_attr {
383 __be64 sys_image_guid;
391 u64 device_cap_flags;
402 int max_qp_init_rd_atom;
403 int max_ee_init_rd_atom;
404 enum ib_atomic_cap atomic_cap;
405 enum ib_atomic_cap masked_atomic_cap;
412 int max_mcast_qp_attach;
413 int max_total_mcast_qp_attach;
418 unsigned int max_fast_reg_page_list_len;
419 unsigned int max_pi_fast_reg_page_list_len;
421 u8 local_ca_ack_delay;
424 struct ib_odp_caps odp_caps;
425 uint64_t timestamp_mask;
426 uint64_t hca_core_clock; /* in KHZ */
427 struct ib_rss_caps rss_caps;
429 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
430 struct ib_tm_caps tm_caps;
431 struct ib_cq_caps cq_caps;
433 /* Max entries for sgl for optimized performance per READ */
450 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
453 case IB_MTU_256: return 256;
454 case IB_MTU_512: return 512;
455 case IB_MTU_1024: return 1024;
456 case IB_MTU_2048: return 2048;
457 case IB_MTU_4096: return 4096;
462 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
466 else if (mtu >= 2048)
468 else if (mtu >= 1024)
476 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
484 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
488 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
491 return OPA_MTU_10240;
492 else if (mtu >= 8192)
495 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
504 IB_PORT_ACTIVE_DEFER = 5
507 enum ib_port_phys_state {
508 IB_PORT_PHYS_STATE_SLEEP = 1,
509 IB_PORT_PHYS_STATE_POLLING = 2,
510 IB_PORT_PHYS_STATE_DISABLED = 3,
511 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
512 IB_PORT_PHYS_STATE_LINK_UP = 5,
513 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
514 IB_PORT_PHYS_STATE_PHY_TEST = 7,
525 static inline int ib_width_enum_to_int(enum ib_port_width width)
528 case IB_WIDTH_1X: return 1;
529 case IB_WIDTH_2X: return 2;
530 case IB_WIDTH_4X: return 4;
531 case IB_WIDTH_8X: return 8;
532 case IB_WIDTH_12X: return 12;
549 IB_STAT_FLAG_OPTIONAL = 1 << 0,
553 * struct rdma_stat_desc
554 * @name - The name of the counter
555 * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
556 * @priv - Driver private information; Core code should not use
558 struct rdma_stat_desc {
565 * struct rdma_hw_stats
566 * @lock - Mutex to protect parallel write access to lifespan and values
567 * of counters, which are 64bits and not guaranteeed to be written
568 * atomicaly on 32bits systems.
569 * @timestamp - Used by the core code to track when the last update was
570 * @lifespan - Used by the core code to determine how old the counters
571 * should be before being updated again. Stored in jiffies, defaults
572 * to 10 milliseconds, drivers can override the default be specifying
573 * their own value during their allocation routine.
574 * @descs - Array of pointers to static descriptors used for the counters
576 * @is_disabled - A bitmap to indicate each counter is currently disabled
578 * @num_counters - How many hardware counters there are. If name is
579 * shorter than this number, a kernel oops will result. Driver authors
580 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
581 * in their code to prevent this.
582 * @value - Array of u64 counters that are accessed by the sysfs code and
583 * filled in by the drivers get_stats routine
585 struct rdma_hw_stats {
586 struct mutex lock; /* Protect lifespan and values[] */
587 unsigned long timestamp;
588 unsigned long lifespan;
589 const struct rdma_stat_desc *descs;
590 unsigned long *is_disabled;
595 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
597 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
598 const struct rdma_stat_desc *descs, int num_counters,
599 unsigned long lifespan);
601 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
603 /* Define bits for the various functionality this port needs to be supported by
606 /* Management 0x00000FFF */
607 #define RDMA_CORE_CAP_IB_MAD 0x00000001
608 #define RDMA_CORE_CAP_IB_SMI 0x00000002
609 #define RDMA_CORE_CAP_IB_CM 0x00000004
610 #define RDMA_CORE_CAP_IW_CM 0x00000008
611 #define RDMA_CORE_CAP_IB_SA 0x00000010
612 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
614 /* Address format 0x000FF000 */
615 #define RDMA_CORE_CAP_AF_IB 0x00001000
616 #define RDMA_CORE_CAP_ETH_AH 0x00002000
617 #define RDMA_CORE_CAP_OPA_AH 0x00004000
618 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
620 /* Protocol 0xFFF00000 */
621 #define RDMA_CORE_CAP_PROT_IB 0x00100000
622 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
623 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
624 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
625 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
626 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
628 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
629 | RDMA_CORE_CAP_PROT_ROCE \
630 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
632 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
633 | RDMA_CORE_CAP_IB_MAD \
634 | RDMA_CORE_CAP_IB_SMI \
635 | RDMA_CORE_CAP_IB_CM \
636 | RDMA_CORE_CAP_IB_SA \
637 | RDMA_CORE_CAP_AF_IB)
638 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
639 | RDMA_CORE_CAP_IB_MAD \
640 | RDMA_CORE_CAP_IB_CM \
641 | RDMA_CORE_CAP_AF_IB \
642 | RDMA_CORE_CAP_ETH_AH)
643 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
644 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
645 | RDMA_CORE_CAP_IB_MAD \
646 | RDMA_CORE_CAP_IB_CM \
647 | RDMA_CORE_CAP_AF_IB \
648 | RDMA_CORE_CAP_ETH_AH)
649 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
650 | RDMA_CORE_CAP_IW_CM)
651 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
652 | RDMA_CORE_CAP_OPA_MAD)
654 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
656 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
658 struct ib_port_attr {
660 enum ib_port_state state;
662 enum ib_mtu active_mtu;
665 unsigned int ip_gids:1;
666 /* This is the value from PortInfo CapabilityMask, defined by IBA */
685 enum ib_device_modify_flags {
686 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
687 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
690 #define IB_DEVICE_NODE_DESC_MAX 64
692 struct ib_device_modify {
694 char node_desc[IB_DEVICE_NODE_DESC_MAX];
697 enum ib_port_modify_flags {
698 IB_PORT_SHUTDOWN = 1,
699 IB_PORT_INIT_TYPE = (1<<2),
700 IB_PORT_RESET_QKEY_CNTR = (1<<3),
701 IB_PORT_OPA_MASK_CHG = (1<<4)
704 struct ib_port_modify {
705 u32 set_port_cap_mask;
706 u32 clr_port_cap_mask;
714 IB_EVENT_QP_ACCESS_ERR,
718 IB_EVENT_PATH_MIG_ERR,
719 IB_EVENT_DEVICE_FATAL,
720 IB_EVENT_PORT_ACTIVE,
723 IB_EVENT_PKEY_CHANGE,
726 IB_EVENT_SRQ_LIMIT_REACHED,
727 IB_EVENT_QP_LAST_WQE_REACHED,
728 IB_EVENT_CLIENT_REREGISTER,
733 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
736 struct ib_device *device;
744 enum ib_event_type event;
747 struct ib_event_handler {
748 struct ib_device *device;
749 void (*handler)(struct ib_event_handler *, struct ib_event *);
750 struct list_head list;
753 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
755 (_ptr)->device = _device; \
756 (_ptr)->handler = _handler; \
757 INIT_LIST_HEAD(&(_ptr)->list); \
760 struct ib_global_route {
761 const struct ib_gid_attr *sgid_attr;
770 __be32 version_tclass_flow;
778 union rdma_network_hdr {
781 /* The IB spec states that if it's IPv4, the header
782 * is located in the last 20 bytes of the header.
785 struct iphdr roce4grh;
789 #define IB_QPN_MASK 0xFFFFFF
792 IB_MULTICAST_QPN = 0xffffff
795 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
796 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
803 IB_RATE_PORT_CURRENT = 0,
804 IB_RATE_2_5_GBPS = 2,
812 IB_RATE_120_GBPS = 10,
813 IB_RATE_14_GBPS = 11,
814 IB_RATE_56_GBPS = 12,
815 IB_RATE_112_GBPS = 13,
816 IB_RATE_168_GBPS = 14,
817 IB_RATE_25_GBPS = 15,
818 IB_RATE_100_GBPS = 16,
819 IB_RATE_200_GBPS = 17,
820 IB_RATE_300_GBPS = 18,
821 IB_RATE_28_GBPS = 19,
822 IB_RATE_50_GBPS = 20,
823 IB_RATE_400_GBPS = 21,
824 IB_RATE_600_GBPS = 22,
828 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
829 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
830 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
831 * @rate: rate to convert.
833 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
836 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
837 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
838 * @rate: rate to convert.
840 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
844 * enum ib_mr_type - memory region type
845 * @IB_MR_TYPE_MEM_REG: memory region that is used for
846 * normal registration
847 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
848 * register any arbitrary sg lists (without
849 * the normal mr constraints - see
851 * @IB_MR_TYPE_DM: memory region that is used for device
852 * memory registration
853 * @IB_MR_TYPE_USER: memory region that is used for the user-space
855 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
856 * without address translations (VA=PA)
857 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
858 * data integrity operations
866 IB_MR_TYPE_INTEGRITY,
869 enum ib_mr_status_check {
870 IB_MR_CHECK_SIG_STATUS = 1,
874 * struct ib_mr_status - Memory region status container
876 * @fail_status: Bitmask of MR checks status. For each
877 * failed check a corresponding status bit is set.
878 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
881 struct ib_mr_status {
883 struct ib_sig_err sig_err;
887 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
889 * @mult: multiple to convert.
891 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
893 struct rdma_ah_init_attr {
894 struct rdma_ah_attr *ah_attr;
896 struct net_device *xmit_slave;
899 enum rdma_ah_attr_type {
900 RDMA_AH_ATTR_TYPE_UNDEFINED,
901 RDMA_AH_ATTR_TYPE_IB,
902 RDMA_AH_ATTR_TYPE_ROCE,
903 RDMA_AH_ATTR_TYPE_OPA,
911 struct roce_ah_attr {
921 struct rdma_ah_attr {
922 struct ib_global_route grh;
927 enum rdma_ah_attr_type type;
929 struct ib_ah_attr ib;
930 struct roce_ah_attr roce;
931 struct opa_ah_attr opa;
939 IB_WC_LOC_EEC_OP_ERR,
944 IB_WC_LOC_ACCESS_ERR,
945 IB_WC_REM_INV_REQ_ERR,
946 IB_WC_REM_ACCESS_ERR,
949 IB_WC_RNR_RETRY_EXC_ERR,
950 IB_WC_LOC_RDD_VIOL_ERR,
951 IB_WC_REM_INV_RD_REQ_ERR,
954 IB_WC_INV_EEC_STATE_ERR,
956 IB_WC_RESP_TIMEOUT_ERR,
960 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
963 IB_WC_SEND = IB_UVERBS_WC_SEND,
964 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
965 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
966 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
967 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
968 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
969 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
970 IB_WC_LSO = IB_UVERBS_WC_TSO,
972 IB_WC_MASKED_COMP_SWAP,
973 IB_WC_MASKED_FETCH_ADD,
975 * Set value of IB_WC_RECV so consumers can test if a completion is a
976 * receive by testing (opcode & IB_WC_RECV).
979 IB_WC_RECV_RDMA_WITH_IMM
984 IB_WC_WITH_IMM = (1<<1),
985 IB_WC_WITH_INVALIDATE = (1<<2),
986 IB_WC_IP_CSUM_OK = (1<<3),
987 IB_WC_WITH_SMAC = (1<<4),
988 IB_WC_WITH_VLAN = (1<<5),
989 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
995 struct ib_cqe *wr_cqe;
997 enum ib_wc_status status;
998 enum ib_wc_opcode opcode;
1004 u32 invalidate_rkey;
1012 u32 port_num; /* valid only for DR SMPs on switches */
1015 u8 network_hdr_type;
1018 enum ib_cq_notify_flags {
1019 IB_CQ_SOLICITED = 1 << 0,
1020 IB_CQ_NEXT_COMP = 1 << 1,
1021 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1022 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1026 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1027 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1028 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1031 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1033 return srq_type == IB_SRQT_XRC ||
1034 srq_type == IB_SRQT_TM;
1037 enum ib_srq_attr_mask {
1038 IB_SRQ_MAX_WR = 1 << 0,
1039 IB_SRQ_LIMIT = 1 << 1,
1042 struct ib_srq_attr {
1048 struct ib_srq_init_attr {
1049 void (*event_handler)(struct ib_event *, void *);
1051 struct ib_srq_attr attr;
1052 enum ib_srq_type srq_type;
1058 struct ib_xrcd *xrcd;
1073 u32 max_inline_data;
1076 * Maximum number of rdma_rw_ctx structures in flight at a time.
1077 * ib_create_qp() will calculate the right amount of neededed WRs
1078 * and MRs based on this.
1090 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1091 * here (and in that order) since the MAD layer uses them as
1092 * indices into a 2-entry table.
1097 IB_QPT_RC = IB_UVERBS_QPT_RC,
1098 IB_QPT_UC = IB_UVERBS_QPT_UC,
1099 IB_QPT_UD = IB_UVERBS_QPT_UD,
1101 IB_QPT_RAW_ETHERTYPE,
1102 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1103 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1104 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1106 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1107 /* Reserve a range for qp types internal to the low level driver.
1108 * These qp types will not be visible at the IB core layer, so the
1109 * IB_QPT_MAX usages should not be affected in the core layer
1111 IB_QPT_RESERVED1 = 0x1000,
1123 enum ib_qp_create_flags {
1124 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1125 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1126 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1127 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1128 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1129 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1130 IB_QP_CREATE_NETIF_QP = 1 << 5,
1131 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1132 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1133 IB_QP_CREATE_SCATTER_FCS =
1134 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1135 IB_QP_CREATE_CVLAN_STRIPPING =
1136 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1137 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1138 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1139 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1140 /* reserve bits 26-31 for low level drivers' internal use */
1141 IB_QP_CREATE_RESERVED_START = 1 << 26,
1142 IB_QP_CREATE_RESERVED_END = 1 << 31,
1146 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1147 * callback to destroy the passed in QP.
1150 struct ib_qp_init_attr {
1151 /* Consumer's event_handler callback must not block */
1152 void (*event_handler)(struct ib_event *, void *);
1155 struct ib_cq *send_cq;
1156 struct ib_cq *recv_cq;
1158 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1159 struct ib_qp_cap cap;
1160 enum ib_sig_type sq_sig_type;
1161 enum ib_qp_type qp_type;
1165 * Only needed for special QP types, or when using the RW API.
1168 struct ib_rwq_ind_table *rwq_ind_tbl;
1172 struct ib_qp_open_attr {
1173 void (*event_handler)(struct ib_event *, void *);
1176 enum ib_qp_type qp_type;
1179 enum ib_rnr_timeout {
1180 IB_RNR_TIMER_655_36 = 0,
1181 IB_RNR_TIMER_000_01 = 1,
1182 IB_RNR_TIMER_000_02 = 2,
1183 IB_RNR_TIMER_000_03 = 3,
1184 IB_RNR_TIMER_000_04 = 4,
1185 IB_RNR_TIMER_000_06 = 5,
1186 IB_RNR_TIMER_000_08 = 6,
1187 IB_RNR_TIMER_000_12 = 7,
1188 IB_RNR_TIMER_000_16 = 8,
1189 IB_RNR_TIMER_000_24 = 9,
1190 IB_RNR_TIMER_000_32 = 10,
1191 IB_RNR_TIMER_000_48 = 11,
1192 IB_RNR_TIMER_000_64 = 12,
1193 IB_RNR_TIMER_000_96 = 13,
1194 IB_RNR_TIMER_001_28 = 14,
1195 IB_RNR_TIMER_001_92 = 15,
1196 IB_RNR_TIMER_002_56 = 16,
1197 IB_RNR_TIMER_003_84 = 17,
1198 IB_RNR_TIMER_005_12 = 18,
1199 IB_RNR_TIMER_007_68 = 19,
1200 IB_RNR_TIMER_010_24 = 20,
1201 IB_RNR_TIMER_015_36 = 21,
1202 IB_RNR_TIMER_020_48 = 22,
1203 IB_RNR_TIMER_030_72 = 23,
1204 IB_RNR_TIMER_040_96 = 24,
1205 IB_RNR_TIMER_061_44 = 25,
1206 IB_RNR_TIMER_081_92 = 26,
1207 IB_RNR_TIMER_122_88 = 27,
1208 IB_RNR_TIMER_163_84 = 28,
1209 IB_RNR_TIMER_245_76 = 29,
1210 IB_RNR_TIMER_327_68 = 30,
1211 IB_RNR_TIMER_491_52 = 31
1214 enum ib_qp_attr_mask {
1216 IB_QP_CUR_STATE = (1<<1),
1217 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1218 IB_QP_ACCESS_FLAGS = (1<<3),
1219 IB_QP_PKEY_INDEX = (1<<4),
1220 IB_QP_PORT = (1<<5),
1221 IB_QP_QKEY = (1<<6),
1223 IB_QP_PATH_MTU = (1<<8),
1224 IB_QP_TIMEOUT = (1<<9),
1225 IB_QP_RETRY_CNT = (1<<10),
1226 IB_QP_RNR_RETRY = (1<<11),
1227 IB_QP_RQ_PSN = (1<<12),
1228 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1229 IB_QP_ALT_PATH = (1<<14),
1230 IB_QP_MIN_RNR_TIMER = (1<<15),
1231 IB_QP_SQ_PSN = (1<<16),
1232 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1233 IB_QP_PATH_MIG_STATE = (1<<18),
1234 IB_QP_CAP = (1<<19),
1235 IB_QP_DEST_QPN = (1<<20),
1236 IB_QP_RESERVED1 = (1<<21),
1237 IB_QP_RESERVED2 = (1<<22),
1238 IB_QP_RESERVED3 = (1<<23),
1239 IB_QP_RESERVED4 = (1<<24),
1240 IB_QP_RATE_LIMIT = (1<<25),
1242 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1267 enum ib_qp_state qp_state;
1268 enum ib_qp_state cur_qp_state;
1269 enum ib_mtu path_mtu;
1270 enum ib_mig_state path_mig_state;
1275 int qp_access_flags;
1276 struct ib_qp_cap cap;
1277 struct rdma_ah_attr ah_attr;
1278 struct rdma_ah_attr alt_ah_attr;
1281 u8 en_sqd_async_notify;
1284 u8 max_dest_rd_atomic;
1293 struct net_device *xmit_slave;
1297 /* These are shared with userspace */
1298 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1299 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1300 IB_WR_SEND = IB_UVERBS_WR_SEND,
1301 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1302 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1303 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1304 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1305 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1306 IB_WR_LSO = IB_UVERBS_WR_TSO,
1307 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1308 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1309 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1310 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1311 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1312 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1313 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1315 /* These are kernel only and can not be issued by userspace */
1316 IB_WR_REG_MR = 0x20,
1317 IB_WR_REG_MR_INTEGRITY,
1319 /* reserve values for low level drivers' internal use.
1320 * These values will not be used at all in the ib core layer.
1322 IB_WR_RESERVED1 = 0xf0,
1334 enum ib_send_flags {
1336 IB_SEND_SIGNALED = (1<<1),
1337 IB_SEND_SOLICITED = (1<<2),
1338 IB_SEND_INLINE = (1<<3),
1339 IB_SEND_IP_CSUM = (1<<4),
1341 /* reserve bits 26-31 for low level drivers' internal use */
1342 IB_SEND_RESERVED_START = (1 << 26),
1343 IB_SEND_RESERVED_END = (1 << 31),
1353 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1357 struct ib_send_wr *next;
1360 struct ib_cqe *wr_cqe;
1362 struct ib_sge *sg_list;
1364 enum ib_wr_opcode opcode;
1368 u32 invalidate_rkey;
1373 struct ib_send_wr wr;
1378 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1380 return container_of(wr, struct ib_rdma_wr, wr);
1383 struct ib_atomic_wr {
1384 struct ib_send_wr wr;
1388 u64 compare_add_mask;
1393 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1395 return container_of(wr, struct ib_atomic_wr, wr);
1399 struct ib_send_wr wr;
1406 u16 pkey_index; /* valid for GSI only */
1407 u32 port_num; /* valid for DR SMPs on switch only */
1410 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1412 return container_of(wr, struct ib_ud_wr, wr);
1416 struct ib_send_wr wr;
1422 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1424 return container_of(wr, struct ib_reg_wr, wr);
1428 struct ib_recv_wr *next;
1431 struct ib_cqe *wr_cqe;
1433 struct ib_sge *sg_list;
1437 enum ib_access_flags {
1438 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1439 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1440 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1441 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1442 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1443 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1444 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1445 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1446 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1448 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1449 IB_ACCESS_SUPPORTED =
1450 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1454 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1455 * are hidden here instead of a uapi header!
1457 enum ib_mr_rereg_flags {
1458 IB_MR_REREG_TRANS = 1,
1459 IB_MR_REREG_PD = (1<<1),
1460 IB_MR_REREG_ACCESS = (1<<2),
1461 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1466 enum rdma_remove_reason {
1468 * Userspace requested uobject deletion or initial try
1469 * to remove uobject via cleanup. Call could fail
1471 RDMA_REMOVE_DESTROY,
1472 /* Context deletion. This call should delete the actual object itself */
1474 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1475 RDMA_REMOVE_DRIVER_REMOVE,
1476 /* uobj is being cleaned-up before being committed */
1478 /* The driver failed to destroy the uobject and is being disconnected */
1479 RDMA_REMOVE_DRIVER_FAILURE,
1482 struct ib_rdmacg_object {
1483 #ifdef CONFIG_CGROUP_RDMA
1484 struct rdma_cgroup *cg; /* owner rdma cgroup */
1488 struct ib_ucontext {
1489 struct ib_device *device;
1490 struct ib_uverbs_file *ufile;
1492 struct ib_rdmacg_object cg_obj;
1494 * Implementation details of the RDMA core, don't use in drivers:
1496 struct rdma_restrack_entry res;
1497 struct xarray mmap_xa;
1501 u64 user_handle; /* handle given to us by userspace */
1502 /* ufile & ucontext owning this object */
1503 struct ib_uverbs_file *ufile;
1504 /* FIXME, save memory: ufile->context == context */
1505 struct ib_ucontext *context; /* associated user context */
1506 void *object; /* containing object */
1507 struct list_head list; /* link to context's list */
1508 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1509 int id; /* index into kernel idr */
1511 atomic_t usecnt; /* protects exclusive access */
1512 struct rcu_head rcu; /* kfree_rcu() overhead */
1514 const struct uverbs_api_object *uapi_object;
1518 const void __user *inbuf;
1519 void __user *outbuf;
1527 struct ib_device *device;
1528 struct ib_uobject *uobject;
1529 atomic_t usecnt; /* count all resources */
1531 u32 unsafe_global_rkey;
1534 * Implementation details of the RDMA core, don't use in drivers:
1536 struct ib_mr *__internal_mr;
1537 struct rdma_restrack_entry res;
1541 struct ib_device *device;
1542 atomic_t usecnt; /* count all exposed resources */
1543 struct inode *inode;
1544 struct rw_semaphore tgt_qps_rwsem;
1545 struct xarray tgt_qps;
1549 struct ib_device *device;
1551 struct ib_uobject *uobject;
1552 const struct ib_gid_attr *sgid_attr;
1553 enum rdma_ah_attr_type type;
1556 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1558 enum ib_poll_context {
1559 IB_POLL_SOFTIRQ, /* poll from softirq context */
1560 IB_POLL_WORKQUEUE, /* poll from workqueue */
1561 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1562 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1564 IB_POLL_DIRECT, /* caller context, no hw completions */
1568 struct ib_device *device;
1569 struct ib_ucq_object *uobject;
1570 ib_comp_handler comp_handler;
1571 void (*event_handler)(struct ib_event *, void *);
1574 unsigned int cqe_used;
1575 atomic_t usecnt; /* count number of work queues */
1576 enum ib_poll_context poll_ctx;
1578 struct list_head pool_entry;
1580 struct irq_poll iop;
1581 struct work_struct work;
1583 struct workqueue_struct *comp_wq;
1586 /* updated only by trace points */
1590 unsigned int comp_vector;
1593 * Implementation details of the RDMA core, don't use in drivers:
1595 struct rdma_restrack_entry res;
1599 struct ib_device *device;
1601 struct ib_usrq_object *uobject;
1602 void (*event_handler)(struct ib_event *, void *);
1604 enum ib_srq_type srq_type;
1611 struct ib_xrcd *xrcd;
1618 * Implementation details of the RDMA core, don't use in drivers:
1620 struct rdma_restrack_entry res;
1623 enum ib_raw_packet_caps {
1624 /* Strip cvlan from incoming packet and report it in the matching work
1625 * completion is supported.
1627 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1628 /* Scatter FCS field of an incoming packet to host memory is supported.
1630 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1631 /* Checksum offloads are supported (for both send and receive). */
1632 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1633 /* When a packet is received for an RQ with no receive WQEs, the
1634 * packet processing is delayed.
1636 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1640 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1650 struct ib_device *device;
1651 struct ib_uwq_object *uobject;
1653 void (*event_handler)(struct ib_event *, void *);
1657 enum ib_wq_state state;
1658 enum ib_wq_type wq_type;
1663 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1664 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1665 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1666 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1667 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1670 struct ib_wq_init_attr {
1672 enum ib_wq_type wq_type;
1676 void (*event_handler)(struct ib_event *, void *);
1677 u32 create_flags; /* Use enum ib_wq_flags */
1680 enum ib_wq_attr_mask {
1681 IB_WQ_STATE = 1 << 0,
1682 IB_WQ_CUR_STATE = 1 << 1,
1683 IB_WQ_FLAGS = 1 << 2,
1687 enum ib_wq_state wq_state;
1688 enum ib_wq_state curr_wq_state;
1689 u32 flags; /* Use enum ib_wq_flags */
1690 u32 flags_mask; /* Use enum ib_wq_flags */
1693 struct ib_rwq_ind_table {
1694 struct ib_device *device;
1695 struct ib_uobject *uobject;
1698 u32 log_ind_tbl_size;
1699 struct ib_wq **ind_tbl;
1702 struct ib_rwq_ind_table_init_attr {
1703 u32 log_ind_tbl_size;
1704 /* Each entry is a pointer to Receive Work Queue */
1705 struct ib_wq **ind_tbl;
1708 enum port_pkey_state {
1709 IB_PORT_PKEY_NOT_VALID = 0,
1710 IB_PORT_PKEY_VALID = 1,
1711 IB_PORT_PKEY_LISTED = 2,
1714 struct ib_qp_security;
1716 struct ib_port_pkey {
1717 enum port_pkey_state state;
1720 struct list_head qp_list;
1721 struct list_head to_error_list;
1722 struct ib_qp_security *sec;
1725 struct ib_ports_pkeys {
1726 struct ib_port_pkey main;
1727 struct ib_port_pkey alt;
1730 struct ib_qp_security {
1732 struct ib_device *dev;
1733 /* Hold this mutex when changing port and pkey settings. */
1735 struct ib_ports_pkeys *ports_pkeys;
1736 /* A list of all open shared QP handles. Required to enforce security
1737 * properly for all users of a shared QP.
1739 struct list_head shared_qp_list;
1742 atomic_t error_list_count;
1743 struct completion error_complete;
1744 int error_comps_pending;
1748 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1749 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1752 struct ib_device *device;
1754 struct ib_cq *send_cq;
1755 struct ib_cq *recv_cq;
1758 struct list_head rdma_mrs;
1759 struct list_head sig_mrs;
1761 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1762 struct list_head xrcd_list;
1764 /* count times opened, mcast attaches, flow attaches */
1766 struct list_head open_list;
1767 struct ib_qp *real_qp;
1768 struct ib_uqp_object *uobject;
1769 void (*event_handler)(struct ib_event *, void *);
1771 /* sgid_attrs associated with the AV's */
1772 const struct ib_gid_attr *av_sgid_attr;
1773 const struct ib_gid_attr *alt_path_sgid_attr;
1777 enum ib_qp_type qp_type;
1778 struct ib_rwq_ind_table *rwq_ind_tbl;
1779 struct ib_qp_security *qp_sec;
1784 * Implementation details of the RDMA core, don't use in drivers:
1786 struct rdma_restrack_entry res;
1788 /* The counter the qp is bind to */
1789 struct rdma_counter *counter;
1793 struct ib_device *device;
1796 struct ib_uobject *uobject;
1801 struct ib_device *device;
1807 unsigned int page_size;
1808 enum ib_mr_type type;
1811 struct ib_uobject *uobject; /* user */
1812 struct list_head qp_entry; /* FR */
1816 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1818 * Implementation details of the RDMA core, don't use in drivers:
1820 struct rdma_restrack_entry res;
1824 struct ib_device *device;
1826 struct ib_uobject *uobject;
1828 enum ib_mw_type type;
1831 /* Supported steering options */
1832 enum ib_flow_attr_type {
1833 /* steering according to rule specifications */
1834 IB_FLOW_ATTR_NORMAL = 0x0,
1835 /* default unicast and multicast rule -
1836 * receive all Eth traffic which isn't steered to any QP
1838 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1839 /* default multicast rule -
1840 * receive all Eth multicast traffic which isn't steered to any QP
1842 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1843 /* sniffer rule - receive all port traffic */
1844 IB_FLOW_ATTR_SNIFFER = 0x3
1847 /* Supported steering header types */
1848 enum ib_flow_spec_type {
1850 IB_FLOW_SPEC_ETH = 0x20,
1851 IB_FLOW_SPEC_IB = 0x22,
1853 IB_FLOW_SPEC_IPV4 = 0x30,
1854 IB_FLOW_SPEC_IPV6 = 0x31,
1855 IB_FLOW_SPEC_ESP = 0x34,
1857 IB_FLOW_SPEC_TCP = 0x40,
1858 IB_FLOW_SPEC_UDP = 0x41,
1859 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1860 IB_FLOW_SPEC_GRE = 0x51,
1861 IB_FLOW_SPEC_MPLS = 0x60,
1862 IB_FLOW_SPEC_INNER = 0x100,
1864 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1865 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1866 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1867 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1869 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1870 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1872 enum ib_flow_flags {
1873 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1874 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1875 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1878 struct ib_flow_eth_filter {
1887 struct ib_flow_spec_eth {
1890 struct ib_flow_eth_filter val;
1891 struct ib_flow_eth_filter mask;
1894 struct ib_flow_ib_filter {
1901 struct ib_flow_spec_ib {
1904 struct ib_flow_ib_filter val;
1905 struct ib_flow_ib_filter mask;
1908 /* IPv4 header flags */
1909 enum ib_ipv4_flags {
1910 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1911 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1912 last have this flag set */
1915 struct ib_flow_ipv4_filter {
1926 struct ib_flow_spec_ipv4 {
1929 struct ib_flow_ipv4_filter val;
1930 struct ib_flow_ipv4_filter mask;
1933 struct ib_flow_ipv6_filter {
1944 struct ib_flow_spec_ipv6 {
1947 struct ib_flow_ipv6_filter val;
1948 struct ib_flow_ipv6_filter mask;
1951 struct ib_flow_tcp_udp_filter {
1958 struct ib_flow_spec_tcp_udp {
1961 struct ib_flow_tcp_udp_filter val;
1962 struct ib_flow_tcp_udp_filter mask;
1965 struct ib_flow_tunnel_filter {
1970 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1971 * the tunnel_id from val has the vni value
1973 struct ib_flow_spec_tunnel {
1976 struct ib_flow_tunnel_filter val;
1977 struct ib_flow_tunnel_filter mask;
1980 struct ib_flow_esp_filter {
1987 struct ib_flow_spec_esp {
1990 struct ib_flow_esp_filter val;
1991 struct ib_flow_esp_filter mask;
1994 struct ib_flow_gre_filter {
1995 __be16 c_ks_res0_ver;
2002 struct ib_flow_spec_gre {
2005 struct ib_flow_gre_filter val;
2006 struct ib_flow_gre_filter mask;
2009 struct ib_flow_mpls_filter {
2015 struct ib_flow_spec_mpls {
2018 struct ib_flow_mpls_filter val;
2019 struct ib_flow_mpls_filter mask;
2022 struct ib_flow_spec_action_tag {
2023 enum ib_flow_spec_type type;
2028 struct ib_flow_spec_action_drop {
2029 enum ib_flow_spec_type type;
2033 struct ib_flow_spec_action_handle {
2034 enum ib_flow_spec_type type;
2036 struct ib_flow_action *act;
2039 enum ib_counters_description {
2044 struct ib_flow_spec_action_count {
2045 enum ib_flow_spec_type type;
2047 struct ib_counters *counters;
2050 union ib_flow_spec {
2055 struct ib_flow_spec_eth eth;
2056 struct ib_flow_spec_ib ib;
2057 struct ib_flow_spec_ipv4 ipv4;
2058 struct ib_flow_spec_tcp_udp tcp_udp;
2059 struct ib_flow_spec_ipv6 ipv6;
2060 struct ib_flow_spec_tunnel tunnel;
2061 struct ib_flow_spec_esp esp;
2062 struct ib_flow_spec_gre gre;
2063 struct ib_flow_spec_mpls mpls;
2064 struct ib_flow_spec_action_tag flow_tag;
2065 struct ib_flow_spec_action_drop drop;
2066 struct ib_flow_spec_action_handle action;
2067 struct ib_flow_spec_action_count flow_count;
2070 struct ib_flow_attr {
2071 enum ib_flow_attr_type type;
2077 union ib_flow_spec flows[];
2082 struct ib_device *device;
2083 struct ib_uobject *uobject;
2086 enum ib_flow_action_type {
2087 IB_FLOW_ACTION_UNSPECIFIED,
2088 IB_FLOW_ACTION_ESP = 1,
2091 struct ib_flow_action_attrs_esp_keymats {
2092 enum ib_uverbs_flow_action_esp_keymat protocol;
2094 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2098 struct ib_flow_action_attrs_esp_replays {
2099 enum ib_uverbs_flow_action_esp_replay protocol;
2101 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2105 enum ib_flow_action_attrs_esp_flags {
2106 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2107 * This is done in order to share the same flags between user-space and
2108 * kernel and spare an unnecessary translation.
2112 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2113 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2116 struct ib_flow_spec_list {
2117 struct ib_flow_spec_list *next;
2118 union ib_flow_spec spec;
2121 struct ib_flow_action_attrs_esp {
2122 struct ib_flow_action_attrs_esp_keymats *keymat;
2123 struct ib_flow_action_attrs_esp_replays *replay;
2124 struct ib_flow_spec_list *encap;
2125 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2126 * Value of 0 is a valid value.
2132 /* Use enum ib_flow_action_attrs_esp_flags */
2134 u64 hard_limit_pkts;
2137 struct ib_flow_action {
2138 struct ib_device *device;
2139 struct ib_uobject *uobject;
2140 enum ib_flow_action_type type;
2146 enum ib_process_mad_flags {
2147 IB_MAD_IGNORE_MKEY = 1,
2148 IB_MAD_IGNORE_BKEY = 2,
2149 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2152 enum ib_mad_result {
2153 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2154 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2155 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2156 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2159 struct ib_port_cache {
2161 struct ib_pkey_cache *pkey;
2162 struct ib_gid_table *gid;
2164 enum ib_port_state port_state;
2167 struct ib_port_immutable {
2174 struct ib_port_data {
2175 struct ib_device *ib_dev;
2177 struct ib_port_immutable immutable;
2179 spinlock_t pkey_list_lock;
2181 spinlock_t netdev_lock;
2183 struct list_head pkey_list;
2185 struct ib_port_cache cache;
2187 struct net_device __rcu *netdev;
2188 struct hlist_node ndev_hash_link;
2189 struct rdma_port_counter port_counter;
2190 struct ib_port *sysfs;
2193 /* rdma netdev type - specifies protocol type */
2194 enum rdma_netdev_t {
2195 RDMA_NETDEV_OPA_VNIC,
2200 * struct rdma_netdev - rdma netdev
2201 * For cases where netstack interfacing is required.
2203 struct rdma_netdev {
2205 struct ib_device *hca;
2210 * cleanup function must be specified.
2211 * FIXME: This is only used for OPA_VNIC and that usage should be
2214 void (*free_rdma_netdev)(struct net_device *netdev);
2216 /* control functions */
2217 void (*set_id)(struct net_device *netdev, int id);
2219 int (*send)(struct net_device *dev, struct sk_buff *skb,
2220 struct ib_ah *address, u32 dqpn);
2222 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2223 union ib_gid *gid, u16 mlid,
2224 int set_qkey, u32 qkey);
2225 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2226 union ib_gid *gid, u16 mlid);
2228 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2231 struct rdma_netdev_alloc_params {
2237 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2238 struct net_device *netdev, void *param);
2241 struct ib_odp_counters {
2243 atomic64_t invalidations;
2244 atomic64_t prefetch;
2247 struct ib_counters {
2248 struct ib_device *device;
2249 struct ib_uobject *uobject;
2250 /* num of objects attached */
2254 struct ib_counters_read_attr {
2257 u32 flags; /* use enum ib_read_counters_flags */
2260 struct uverbs_attr_bundle;
2262 struct iw_cm_conn_param;
2264 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2265 .size_##ib_struct = \
2266 (sizeof(struct drv_struct) + \
2267 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2268 BUILD_BUG_ON_ZERO( \
2269 !__same_type(((struct drv_struct *)NULL)->member, \
2272 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2273 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2276 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2277 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2280 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2281 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2283 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2285 struct rdma_user_mmap_entry {
2287 struct ib_ucontext *ucontext;
2288 unsigned long start_pgoff;
2290 bool driver_removed;
2293 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2295 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2297 return (u64)entry->start_pgoff << PAGE_SHIFT;
2301 * struct ib_device_ops - InfiniBand device operations
2302 * This structure defines all the InfiniBand device operations, providers will
2303 * need to define the supported operations, otherwise they will be set to null.
2305 struct ib_device_ops {
2306 struct module *owner;
2307 enum rdma_driver_id driver_id;
2309 unsigned int uverbs_no_driver_id_binding:1;
2312 * NOTE: New drivers should not make use of device_group; instead new
2313 * device parameter should be exposed via netlink command. This
2314 * mechanism exists only for existing drivers.
2316 const struct attribute_group *device_group;
2317 const struct attribute_group **port_groups;
2319 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2320 const struct ib_send_wr **bad_send_wr);
2321 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2322 const struct ib_recv_wr **bad_recv_wr);
2323 void (*drain_rq)(struct ib_qp *qp);
2324 void (*drain_sq)(struct ib_qp *qp);
2325 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2326 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2327 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2328 int (*post_srq_recv)(struct ib_srq *srq,
2329 const struct ib_recv_wr *recv_wr,
2330 const struct ib_recv_wr **bad_recv_wr);
2331 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2332 u32 port_num, const struct ib_wc *in_wc,
2333 const struct ib_grh *in_grh,
2334 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2335 size_t *out_mad_size, u16 *out_mad_pkey_index);
2336 int (*query_device)(struct ib_device *device,
2337 struct ib_device_attr *device_attr,
2338 struct ib_udata *udata);
2339 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2340 struct ib_device_modify *device_modify);
2341 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2342 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2344 int (*query_port)(struct ib_device *device, u32 port_num,
2345 struct ib_port_attr *port_attr);
2346 int (*modify_port)(struct ib_device *device, u32 port_num,
2347 int port_modify_mask,
2348 struct ib_port_modify *port_modify);
2350 * The following mandatory functions are used only at device
2351 * registration. Keep functions such as these at the end of this
2352 * structure to avoid cache line misses when accessing struct ib_device
2355 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2356 struct ib_port_immutable *immutable);
2357 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2360 * When calling get_netdev, the HW vendor's driver should return the
2361 * net device of device @device at port @port_num or NULL if such
2362 * a net device doesn't exist. The vendor driver should call dev_hold
2363 * on this net device. The HW vendor's device driver must guarantee
2364 * that this function returns NULL before the net device has finished
2365 * NETDEV_UNREGISTER state.
2367 struct net_device *(*get_netdev)(struct ib_device *device,
2370 * rdma netdev operation
2372 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2373 * must return -EOPNOTSUPP if it doesn't support the specified type.
2375 struct net_device *(*alloc_rdma_netdev)(
2376 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2377 const char *name, unsigned char name_assign_type,
2378 void (*setup)(struct net_device *));
2380 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2381 enum rdma_netdev_t type,
2382 struct rdma_netdev_alloc_params *params);
2384 * query_gid should be return GID value for @device, when @port_num
2385 * link layer is either IB or iWarp. It is no-op if @port_num port
2386 * is RoCE link layer.
2388 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2391 * When calling add_gid, the HW vendor's driver should add the gid
2392 * of device of port at gid index available at @attr. Meta-info of
2393 * that gid (for example, the network device related to this gid) is
2394 * available at @attr. @context allows the HW vendor driver to store
2395 * extra information together with a GID entry. The HW vendor driver may
2396 * allocate memory to contain this information and store it in @context
2397 * when a new GID entry is written to. Params are consistent until the
2398 * next call of add_gid or delete_gid. The function should return 0 on
2399 * success or error otherwise. The function could be called
2400 * concurrently for different ports. This function is only called when
2401 * roce_gid_table is used.
2403 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2405 * When calling del_gid, the HW vendor's driver should delete the
2406 * gid of device @device at gid index gid_index of port port_num
2407 * available in @attr.
2408 * Upon the deletion of a GID entry, the HW vendor must free any
2409 * allocated memory. The caller will clear @context afterwards.
2410 * This function is only called when roce_gid_table is used.
2412 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2413 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2415 int (*alloc_ucontext)(struct ib_ucontext *context,
2416 struct ib_udata *udata);
2417 void (*dealloc_ucontext)(struct ib_ucontext *context);
2418 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2420 * This will be called once refcount of an entry in mmap_xa reaches
2421 * zero. The type of the memory that was mapped may differ between
2422 * entries and is opaque to the rdma_user_mmap interface.
2423 * Therefore needs to be implemented by the driver in mmap_free.
2425 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2426 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2427 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2428 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2429 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2430 struct ib_udata *udata);
2431 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2432 struct ib_udata *udata);
2433 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2434 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2435 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2436 int (*create_srq)(struct ib_srq *srq,
2437 struct ib_srq_init_attr *srq_init_attr,
2438 struct ib_udata *udata);
2439 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2440 enum ib_srq_attr_mask srq_attr_mask,
2441 struct ib_udata *udata);
2442 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2443 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2444 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2445 struct ib_udata *udata);
2446 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2447 int qp_attr_mask, struct ib_udata *udata);
2448 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2449 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2450 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2451 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2452 struct ib_udata *udata);
2453 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2454 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2455 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2456 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2457 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2458 u64 virt_addr, int mr_access_flags,
2459 struct ib_udata *udata);
2460 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2461 u64 length, u64 virt_addr, int fd,
2462 int mr_access_flags,
2463 struct ib_udata *udata);
2464 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2465 u64 length, u64 virt_addr,
2466 int mr_access_flags, struct ib_pd *pd,
2467 struct ib_udata *udata);
2468 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2469 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2471 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2472 u32 max_num_data_sg,
2473 u32 max_num_meta_sg);
2474 int (*advise_mr)(struct ib_pd *pd,
2475 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2476 struct ib_sge *sg_list, u32 num_sge,
2477 struct uverbs_attr_bundle *attrs);
2480 * Kernel users should universally support relaxed ordering (RO), as
2481 * they are designed to read data only after observing the CQE and use
2482 * the DMA API correctly.
2484 * Some drivers implicitly enable RO if platform supports it.
2486 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2487 unsigned int *sg_offset);
2488 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2489 struct ib_mr_status *mr_status);
2490 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2491 int (*dealloc_mw)(struct ib_mw *mw);
2492 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2493 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2494 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2495 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2496 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2497 struct ib_flow_attr *flow_attr,
2498 struct ib_udata *udata);
2499 int (*destroy_flow)(struct ib_flow *flow_id);
2500 struct ib_flow_action *(*create_flow_action_esp)(
2501 struct ib_device *device,
2502 const struct ib_flow_action_attrs_esp *attr,
2503 struct uverbs_attr_bundle *attrs);
2504 int (*destroy_flow_action)(struct ib_flow_action *action);
2505 int (*modify_flow_action_esp)(
2506 struct ib_flow_action *action,
2507 const struct ib_flow_action_attrs_esp *attr,
2508 struct uverbs_attr_bundle *attrs);
2509 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2511 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2512 struct ifla_vf_info *ivf);
2513 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2514 struct ifla_vf_stats *stats);
2515 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2516 struct ifla_vf_guid *node_guid,
2517 struct ifla_vf_guid *port_guid);
2518 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2520 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2521 struct ib_wq_init_attr *init_attr,
2522 struct ib_udata *udata);
2523 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2524 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2525 u32 wq_attr_mask, struct ib_udata *udata);
2526 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2527 struct ib_rwq_ind_table_init_attr *init_attr,
2528 struct ib_udata *udata);
2529 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2530 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2531 struct ib_ucontext *context,
2532 struct ib_dm_alloc_attr *attr,
2533 struct uverbs_attr_bundle *attrs);
2534 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2535 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2536 struct ib_dm_mr_attr *attr,
2537 struct uverbs_attr_bundle *attrs);
2538 int (*create_counters)(struct ib_counters *counters,
2539 struct uverbs_attr_bundle *attrs);
2540 int (*destroy_counters)(struct ib_counters *counters);
2541 int (*read_counters)(struct ib_counters *counters,
2542 struct ib_counters_read_attr *counters_read_attr,
2543 struct uverbs_attr_bundle *attrs);
2544 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2545 int data_sg_nents, unsigned int *data_sg_offset,
2546 struct scatterlist *meta_sg, int meta_sg_nents,
2547 unsigned int *meta_sg_offset);
2550 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2551 * fill in the driver initialized data. The struct is kfree()'ed by
2552 * the sysfs core when the device is removed. A lifespan of -1 in the
2553 * return struct tells the core to set a default lifespan.
2555 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2556 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2559 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2560 * @index - The index in the value array we wish to have updated, or
2561 * num_counters if we want all stats updated
2563 * < 0 - Error, no counters updated
2564 * index - Updated the single counter pointed to by index
2565 * num_counters - Updated all counters (will reset the timestamp
2566 * and prevent further calls for lifespan milliseconds)
2567 * Drivers are allowed to update all counters in leiu of just the
2568 * one given in index at their option
2570 int (*get_hw_stats)(struct ib_device *device,
2571 struct rdma_hw_stats *stats, u32 port, int index);
2574 * modify_hw_stat - Modify the counter configuration
2575 * @enable: true/false when enable/disable a counter
2576 * Return codes - 0 on success or error code otherwise.
2578 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2579 unsigned int counter_index, bool enable);
2581 * Allows rdma drivers to add their own restrack attributes.
2583 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2584 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2585 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2586 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2587 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2588 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2589 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2591 /* Device lifecycle callbacks */
2593 * Called after the device becomes registered, before clients are
2596 int (*enable_driver)(struct ib_device *dev);
2598 * This is called as part of ib_dealloc_device().
2600 void (*dealloc_driver)(struct ib_device *dev);
2602 /* iWarp CM callbacks */
2603 void (*iw_add_ref)(struct ib_qp *qp);
2604 void (*iw_rem_ref)(struct ib_qp *qp);
2605 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2606 int (*iw_connect)(struct iw_cm_id *cm_id,
2607 struct iw_cm_conn_param *conn_param);
2608 int (*iw_accept)(struct iw_cm_id *cm_id,
2609 struct iw_cm_conn_param *conn_param);
2610 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2612 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2613 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2615 * counter_bind_qp - Bind a QP to a counter.
2616 * @counter - The counter to be bound. If counter->id is zero then
2617 * the driver needs to allocate a new counter and set counter->id
2619 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2621 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2622 * counter and bind it onto the default one
2624 int (*counter_unbind_qp)(struct ib_qp *qp);
2626 * counter_dealloc -De-allocate the hw counter
2628 int (*counter_dealloc)(struct rdma_counter *counter);
2630 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2631 * the driver initialized data.
2633 struct rdma_hw_stats *(*counter_alloc_stats)(
2634 struct rdma_counter *counter);
2636 * counter_update_stats - Query the stats value of this counter
2638 int (*counter_update_stats)(struct rdma_counter *counter);
2641 * Allows rdma drivers to add their own restrack attributes
2642 * dumped via 'rdma stat' iproute2 command.
2644 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2646 /* query driver for its ucontext properties */
2647 int (*query_ucontext)(struct ib_ucontext *context,
2648 struct uverbs_attr_bundle *attrs);
2651 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2652 * Everyone else relies on Linux memory management model.
2654 int (*get_numa_node)(struct ib_device *dev);
2656 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2657 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2658 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2659 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2660 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2661 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2662 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2663 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2664 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2665 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2668 struct ib_core_device {
2669 /* device must be the first element in structure until,
2670 * union of ib_core_device and device exists in ib_device.
2673 possible_net_t rdma_net;
2674 struct kobject *ports_kobj;
2675 struct list_head port_list;
2676 struct ib_device *owner; /* reach back to owner ib_device */
2679 struct rdma_restrack_root;
2681 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2682 struct device *dma_device;
2683 struct ib_device_ops ops;
2684 char name[IB_DEVICE_NAME_MAX];
2685 struct rcu_head rcu_head;
2687 struct list_head event_handler_list;
2688 /* Protects event_handler_list */
2689 struct rw_semaphore event_handler_rwsem;
2691 /* Protects QP's event_handler calls and open_qp list */
2692 spinlock_t qp_open_list_lock;
2694 struct rw_semaphore client_data_rwsem;
2695 struct xarray client_data;
2696 struct mutex unregistration_lock;
2698 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2699 rwlock_t cache_lock;
2701 * port_data is indexed by port number
2703 struct ib_port_data *port_data;
2705 int num_comp_vectors;
2709 struct ib_core_device coredev;
2712 /* First group is for device attributes,
2713 * Second group is for driver provided attributes (optional).
2714 * Third group is for the hw_stats
2715 * It is a NULL terminated array.
2717 const struct attribute_group *groups[4];
2719 u64 uverbs_cmd_mask;
2721 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2725 /* Indicates kernel verbs support, should not be used in drivers */
2726 u16 kverbs_provider:1;
2727 /* CQ adaptive moderation (RDMA DIM) */
2731 struct ib_device_attr attrs;
2732 struct hw_stats_device_data *hw_stats_data;
2734 #ifdef CONFIG_CGROUP_RDMA
2735 struct rdmacg_device cg_device;
2740 spinlock_t cq_pools_lock;
2741 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2743 struct rdma_restrack_root *res;
2745 const struct uapi_definition *driver_def;
2748 * Positive refcount indicates that the device is currently
2749 * registered and cannot be unregistered.
2751 refcount_t refcount;
2752 struct completion unreg_completion;
2753 struct work_struct unregistration_work;
2755 const struct rdma_link_ops *link_ops;
2757 /* Protects compat_devs xarray modifications */
2758 struct mutex compat_devs_mutex;
2759 /* Maintains compat devices for each net namespace */
2760 struct xarray compat_devs;
2762 /* Used by iWarp CM */
2763 char iw_ifname[IFNAMSIZ];
2764 u32 iw_driver_flags;
2768 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2769 gfp_t gfp, bool is_numa_aware)
2771 if (is_numa_aware && dev->ops.get_numa_node)
2772 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2774 return kzalloc(size, gfp);
2777 struct ib_client_nl_info;
2780 int (*add)(struct ib_device *ibdev);
2781 void (*remove)(struct ib_device *, void *client_data);
2782 void (*rename)(struct ib_device *dev, void *client_data);
2783 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2784 struct ib_client_nl_info *res);
2785 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2787 /* Returns the net_dev belonging to this ib_client and matching the
2789 * @dev: An RDMA device that the net_dev use for communication.
2790 * @port: A physical port number on the RDMA device.
2791 * @pkey: P_Key that the net_dev uses if applicable.
2792 * @gid: A GID that the net_dev uses to communicate.
2793 * @addr: An IP address the net_dev is configured with.
2794 * @client_data: The device's client data set by ib_set_client_data().
2796 * An ib_client that implements a net_dev on top of RDMA devices
2797 * (such as IP over IB) should implement this callback, allowing the
2798 * rdma_cm module to find the right net_dev for a given request.
2800 * The caller is responsible for calling dev_put on the returned
2802 struct net_device *(*get_net_dev_by_params)(
2803 struct ib_device *dev,
2806 const union ib_gid *gid,
2807 const struct sockaddr *addr,
2811 struct completion uses_zero;
2814 /* kverbs are not required by the client */
2819 * IB block DMA iterator
2821 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2822 * to a HW supported page size.
2824 struct ib_block_iter {
2825 /* internal states */
2826 struct scatterlist *__sg; /* sg holding the current aligned block */
2827 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2828 unsigned int __sg_nents; /* number of SG entries */
2829 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2830 unsigned int __pg_bit; /* alignment of current block */
2833 struct ib_device *_ib_alloc_device(size_t size);
2834 #define ib_alloc_device(drv_struct, member) \
2835 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2836 BUILD_BUG_ON_ZERO(offsetof( \
2837 struct drv_struct, member))), \
2838 struct drv_struct, member)
2840 void ib_dealloc_device(struct ib_device *device);
2842 void ib_get_device_fw_str(struct ib_device *device, char *str);
2844 int ib_register_device(struct ib_device *device, const char *name,
2845 struct device *dma_device);
2846 void ib_unregister_device(struct ib_device *device);
2847 void ib_unregister_driver(enum rdma_driver_id driver_id);
2848 void ib_unregister_device_and_put(struct ib_device *device);
2849 void ib_unregister_device_queued(struct ib_device *ib_dev);
2851 int ib_register_client (struct ib_client *client);
2852 void ib_unregister_client(struct ib_client *client);
2854 void __rdma_block_iter_start(struct ib_block_iter *biter,
2855 struct scatterlist *sglist,
2857 unsigned long pgsz);
2858 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2861 * rdma_block_iter_dma_address - get the aligned dma address of the current
2862 * block held by the block iterator.
2863 * @biter: block iterator holding the memory block
2865 static inline dma_addr_t
2866 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2868 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2872 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2873 * @sglist: sglist to iterate over
2874 * @biter: block iterator holding the memory block
2875 * @nents: maximum number of sg entries to iterate over
2876 * @pgsz: best HW supported page size to use
2878 * Callers may use rdma_block_iter_dma_address() to get each
2879 * blocks aligned DMA address.
2881 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2882 for (__rdma_block_iter_start(biter, sglist, nents, \
2884 __rdma_block_iter_next(biter);)
2887 * ib_get_client_data - Get IB client context
2888 * @device:Device to get context for
2889 * @client:Client to get context for
2891 * ib_get_client_data() returns the client context data set with
2892 * ib_set_client_data(). This can only be called while the client is
2893 * registered to the device, once the ib_client remove() callback returns this
2896 static inline void *ib_get_client_data(struct ib_device *device,
2897 struct ib_client *client)
2899 return xa_load(&device->client_data, client->client_id);
2901 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2903 void ib_set_device_ops(struct ib_device *device,
2904 const struct ib_device_ops *ops);
2906 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2907 unsigned long pfn, unsigned long size, pgprot_t prot,
2908 struct rdma_user_mmap_entry *entry);
2909 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2910 struct rdma_user_mmap_entry *entry,
2912 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2913 struct rdma_user_mmap_entry *entry,
2914 size_t length, u32 min_pgoff,
2918 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2919 struct rdma_user_mmap_entry *entry,
2920 size_t length, u32 pgoff)
2922 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2926 struct rdma_user_mmap_entry *
2927 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2928 unsigned long pgoff);
2929 struct rdma_user_mmap_entry *
2930 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2931 struct vm_area_struct *vma);
2932 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2934 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2936 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2938 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2941 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2943 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2946 static inline bool ib_is_buffer_cleared(const void __user *p,
2952 if (len > USHRT_MAX)
2955 buf = memdup_user(p, len);
2959 ret = !memchr_inv(buf, 0, len);
2964 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2968 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2972 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2973 * contains all required attributes and no attributes not allowed for
2974 * the given QP state transition.
2975 * @cur_state: Current QP state
2976 * @next_state: Next QP state
2978 * @mask: Mask of supplied QP attributes
2980 * This function is a helper function that a low-level driver's
2981 * modify_qp method can use to validate the consumer's input. It
2982 * checks that cur_state and next_state are valid QP states, that a
2983 * transition from cur_state to next_state is allowed by the IB spec,
2984 * and that the attribute mask supplied is allowed for the transition.
2986 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2987 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2989 void ib_register_event_handler(struct ib_event_handler *event_handler);
2990 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2991 void ib_dispatch_event(const struct ib_event *event);
2993 int ib_query_port(struct ib_device *device,
2994 u32 port_num, struct ib_port_attr *port_attr);
2996 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3000 * rdma_cap_ib_switch - Check if the device is IB switch
3001 * @device: Device to check
3003 * Device driver is responsible for setting is_switch bit on
3004 * in ib_device structure at init time.
3006 * Return: true if the device is IB switch.
3008 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3010 return device->is_switch;
3014 * rdma_start_port - Return the first valid port number for the device
3017 * @device: Device to be checked
3019 * Return start port number
3021 static inline u32 rdma_start_port(const struct ib_device *device)
3023 return rdma_cap_ib_switch(device) ? 0 : 1;
3027 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3028 * @device - The struct ib_device * to iterate over
3029 * @iter - The unsigned int to store the port number
3031 #define rdma_for_each_port(device, iter) \
3032 for (iter = rdma_start_port(device + \
3033 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3035 iter <= rdma_end_port(device); iter++)
3038 * rdma_end_port - Return the last valid port number for the device
3041 * @device: Device to be checked
3043 * Return last port number
3045 static inline u32 rdma_end_port(const struct ib_device *device)
3047 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3050 static inline int rdma_is_port_valid(const struct ib_device *device,
3053 return (port >= rdma_start_port(device) &&
3054 port <= rdma_end_port(device));
3057 static inline bool rdma_is_grh_required(const struct ib_device *device,
3060 return device->port_data[port_num].immutable.core_cap_flags &
3061 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3064 static inline bool rdma_protocol_ib(const struct ib_device *device,
3067 return device->port_data[port_num].immutable.core_cap_flags &
3068 RDMA_CORE_CAP_PROT_IB;
3071 static inline bool rdma_protocol_roce(const struct ib_device *device,
3074 return device->port_data[port_num].immutable.core_cap_flags &
3075 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3078 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3081 return device->port_data[port_num].immutable.core_cap_flags &
3082 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3085 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3088 return device->port_data[port_num].immutable.core_cap_flags &
3089 RDMA_CORE_CAP_PROT_ROCE;
3092 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3095 return device->port_data[port_num].immutable.core_cap_flags &
3096 RDMA_CORE_CAP_PROT_IWARP;
3099 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3102 return rdma_protocol_ib(device, port_num) ||
3103 rdma_protocol_roce(device, port_num);
3106 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3109 return device->port_data[port_num].immutable.core_cap_flags &
3110 RDMA_CORE_CAP_PROT_RAW_PACKET;
3113 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3116 return device->port_data[port_num].immutable.core_cap_flags &
3117 RDMA_CORE_CAP_PROT_USNIC;
3121 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3122 * Management Datagrams.
3123 * @device: Device to check
3124 * @port_num: Port number to check
3126 * Management Datagrams (MAD) are a required part of the InfiniBand
3127 * specification and are supported on all InfiniBand devices. A slightly
3128 * extended version are also supported on OPA interfaces.
3130 * Return: true if the port supports sending/receiving of MAD packets.
3132 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3134 return device->port_data[port_num].immutable.core_cap_flags &
3135 RDMA_CORE_CAP_IB_MAD;
3139 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3140 * Management Datagrams.
3141 * @device: Device to check
3142 * @port_num: Port number to check
3144 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3145 * datagrams with their own versions. These OPA MADs share many but not all of
3146 * the characteristics of InfiniBand MADs.
3148 * OPA MADs differ in the following ways:
3150 * 1) MADs are variable size up to 2K
3151 * IBTA defined MADs remain fixed at 256 bytes
3152 * 2) OPA SMPs must carry valid PKeys
3153 * 3) OPA SMP packets are a different format
3155 * Return: true if the port supports OPA MAD packet formats.
3157 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3159 return device->port_data[port_num].immutable.core_cap_flags &
3160 RDMA_CORE_CAP_OPA_MAD;
3164 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3165 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3166 * @device: Device to check
3167 * @port_num: Port number to check
3169 * Each InfiniBand node is required to provide a Subnet Management Agent
3170 * that the subnet manager can access. Prior to the fabric being fully
3171 * configured by the subnet manager, the SMA is accessed via a well known
3172 * interface called the Subnet Management Interface (SMI). This interface
3173 * uses directed route packets to communicate with the SM to get around the
3174 * chicken and egg problem of the SM needing to know what's on the fabric
3175 * in order to configure the fabric, and needing to configure the fabric in
3176 * order to send packets to the devices on the fabric. These directed
3177 * route packets do not need the fabric fully configured in order to reach
3178 * their destination. The SMI is the only method allowed to send
3179 * directed route packets on an InfiniBand fabric.
3181 * Return: true if the port provides an SMI.
3183 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3185 return device->port_data[port_num].immutable.core_cap_flags &
3186 RDMA_CORE_CAP_IB_SMI;
3190 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3191 * Communication Manager.
3192 * @device: Device to check
3193 * @port_num: Port number to check
3195 * The InfiniBand Communication Manager is one of many pre-defined General
3196 * Service Agents (GSA) that are accessed via the General Service
3197 * Interface (GSI). It's role is to facilitate establishment of connections
3198 * between nodes as well as other management related tasks for established
3201 * Return: true if the port supports an IB CM (this does not guarantee that
3202 * a CM is actually running however).
3204 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3206 return device->port_data[port_num].immutable.core_cap_flags &
3207 RDMA_CORE_CAP_IB_CM;
3211 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3212 * Communication Manager.
3213 * @device: Device to check
3214 * @port_num: Port number to check
3216 * Similar to above, but specific to iWARP connections which have a different
3217 * managment protocol than InfiniBand.
3219 * Return: true if the port supports an iWARP CM (this does not guarantee that
3220 * a CM is actually running however).
3222 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3224 return device->port_data[port_num].immutable.core_cap_flags &
3225 RDMA_CORE_CAP_IW_CM;
3229 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3230 * Subnet Administration.
3231 * @device: Device to check
3232 * @port_num: Port number to check
3234 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3235 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3236 * fabrics, devices should resolve routes to other hosts by contacting the
3237 * SA to query the proper route.
3239 * Return: true if the port should act as a client to the fabric Subnet
3240 * Administration interface. This does not imply that the SA service is
3243 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3245 return device->port_data[port_num].immutable.core_cap_flags &
3246 RDMA_CORE_CAP_IB_SA;
3250 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3252 * @device: Device to check
3253 * @port_num: Port number to check
3255 * InfiniBand multicast registration is more complex than normal IPv4 or
3256 * IPv6 multicast registration. Each Host Channel Adapter must register
3257 * with the Subnet Manager when it wishes to join a multicast group. It
3258 * should do so only once regardless of how many queue pairs it subscribes
3259 * to this group. And it should leave the group only after all queue pairs
3260 * attached to the group have been detached.
3262 * Return: true if the port must undertake the additional adminstrative
3263 * overhead of registering/unregistering with the SM and tracking of the
3264 * total number of queue pairs attached to the multicast group.
3266 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3269 return rdma_cap_ib_sa(device, port_num);
3273 * rdma_cap_af_ib - Check if the port of device has the capability
3274 * Native Infiniband Address.
3275 * @device: Device to check
3276 * @port_num: Port number to check
3278 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3279 * GID. RoCE uses a different mechanism, but still generates a GID via
3280 * a prescribed mechanism and port specific data.
3282 * Return: true if the port uses a GID address to identify devices on the
3285 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3287 return device->port_data[port_num].immutable.core_cap_flags &
3288 RDMA_CORE_CAP_AF_IB;
3292 * rdma_cap_eth_ah - Check if the port of device has the capability
3293 * Ethernet Address Handle.
3294 * @device: Device to check
3295 * @port_num: Port number to check
3297 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3298 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3299 * port. Normally, packet headers are generated by the sending host
3300 * adapter, but when sending connectionless datagrams, we must manually
3301 * inject the proper headers for the fabric we are communicating over.
3303 * Return: true if we are running as a RoCE port and must force the
3304 * addition of a Global Route Header built from our Ethernet Address
3305 * Handle into our header list for connectionless packets.
3307 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3309 return device->port_data[port_num].immutable.core_cap_flags &
3310 RDMA_CORE_CAP_ETH_AH;
3314 * rdma_cap_opa_ah - Check if the port of device supports
3315 * OPA Address handles
3316 * @device: Device to check
3317 * @port_num: Port number to check
3319 * Return: true if we are running on an OPA device which supports
3320 * the extended OPA addressing.
3322 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3324 return (device->port_data[port_num].immutable.core_cap_flags &
3325 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3329 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3332 * @port_num: Port number
3334 * This MAD size includes the MAD headers and MAD payload. No other headers
3337 * Return the max MAD size required by the Port. Will return 0 if the port
3338 * does not support MADs
3340 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3343 return device->port_data[port_num].immutable.max_mad_size;
3347 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3348 * @device: Device to check
3349 * @port_num: Port number to check
3351 * RoCE GID table mechanism manages the various GIDs for a device.
3353 * NOTE: if allocating the port's GID table has failed, this call will still
3354 * return true, but any RoCE GID table API will fail.
3356 * Return: true if the port uses RoCE GID table mechanism in order to manage
3359 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3362 return rdma_protocol_roce(device, port_num) &&
3363 device->ops.add_gid && device->ops.del_gid;
3367 * Check if the device supports READ W/ INVALIDATE.
3369 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3372 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3373 * has support for it yet.
3375 return rdma_protocol_iwarp(dev, port_num);
3379 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3381 * @port_num: 1 based Port number
3383 * Return true if port is an Intel OPA port , false if not
3385 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3388 return (device->port_data[port_num].immutable.core_cap_flags &
3389 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3393 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3395 * @port_num: Port number
3396 * @mtu: enum value of MTU
3398 * Return the MTU size supported by the port as an integer value. Will return
3399 * -1 if enum value of mtu is not supported.
3401 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3404 if (rdma_core_cap_opa_port(device, port))
3405 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3407 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3411 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3413 * @port_num: Port number
3414 * @attr: port attribute
3416 * Return the MTU size supported by the port as an integer value.
3418 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3419 struct ib_port_attr *attr)
3421 if (rdma_core_cap_opa_port(device, port))
3422 return attr->phys_mtu;
3424 return ib_mtu_enum_to_int(attr->max_mtu);
3427 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3429 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3430 struct ifla_vf_info *info);
3431 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3432 struct ifla_vf_stats *stats);
3433 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3434 struct ifla_vf_guid *node_guid,
3435 struct ifla_vf_guid *port_guid);
3436 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3439 int ib_query_pkey(struct ib_device *device,
3440 u32 port_num, u16 index, u16 *pkey);
3442 int ib_modify_device(struct ib_device *device,
3443 int device_modify_mask,
3444 struct ib_device_modify *device_modify);
3446 int ib_modify_port(struct ib_device *device,
3447 u32 port_num, int port_modify_mask,
3448 struct ib_port_modify *port_modify);
3450 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3451 u32 *port_num, u16 *index);
3453 int ib_find_pkey(struct ib_device *device,
3454 u32 port_num, u16 pkey, u16 *index);
3458 * Create a memory registration for all memory in the system and place
3459 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3460 * ULPs to avoid the overhead of dynamic MRs.
3462 * This flag is generally considered unsafe and must only be used in
3463 * extremly trusted environments. Every use of it will log a warning
3464 * in the kernel log.
3466 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3469 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3470 const char *caller);
3473 * ib_alloc_pd - Allocates an unused protection domain.
3474 * @device: The device on which to allocate the protection domain.
3475 * @flags: protection domain flags
3477 * A protection domain object provides an association between QPs, shared
3478 * receive queues, address handles, memory regions, and memory windows.
3480 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3481 * memory operations.
3483 #define ib_alloc_pd(device, flags) \
3484 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3486 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3489 * ib_dealloc_pd - Deallocate kernel PD
3490 * @pd: The protection domain
3492 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3494 static inline void ib_dealloc_pd(struct ib_pd *pd)
3496 int ret = ib_dealloc_pd_user(pd, NULL);
3498 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3501 enum rdma_create_ah_flags {
3502 /* In a sleepable context */
3503 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3507 * rdma_create_ah - Creates an address handle for the given address vector.
3508 * @pd: The protection domain associated with the address handle.
3509 * @ah_attr: The attributes of the address vector.
3510 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3512 * The address handle is used to reference a local or global destination
3513 * in all UD QP post sends.
3515 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3519 * rdma_create_user_ah - Creates an address handle for the given address vector.
3520 * It resolves destination mac address for ah attribute of RoCE type.
3521 * @pd: The protection domain associated with the address handle.
3522 * @ah_attr: The attributes of the address vector.
3523 * @udata: pointer to user's input output buffer information need by
3526 * It returns 0 on success and returns appropriate error code on error.
3527 * The address handle is used to reference a local or global destination
3528 * in all UD QP post sends.
3530 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3531 struct rdma_ah_attr *ah_attr,
3532 struct ib_udata *udata);
3534 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3536 * @hdr: the L3 header to parse
3537 * @net_type: type of header to parse
3538 * @sgid: place to store source gid
3539 * @dgid: place to store destination gid
3541 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3542 enum rdma_network_type net_type,
3543 union ib_gid *sgid, union ib_gid *dgid);
3546 * ib_get_rdma_header_version - Get the header version
3547 * @hdr: the L3 header to parse
3549 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3552 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3554 * @device: Device on which the received message arrived.
3555 * @port_num: Port on which the received message arrived.
3556 * @wc: Work completion associated with the received message.
3557 * @grh: References the received global route header. This parameter is
3558 * ignored unless the work completion indicates that the GRH is valid.
3559 * @ah_attr: Returned attributes that can be used when creating an address
3560 * handle for replying to the message.
3561 * When ib_init_ah_attr_from_wc() returns success,
3562 * (a) for IB link layer it optionally contains a reference to SGID attribute
3563 * when GRH is present for IB link layer.
3564 * (b) for RoCE link layer it contains a reference to SGID attribute.
3565 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3566 * attributes which are initialized using ib_init_ah_attr_from_wc().
3569 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3570 const struct ib_wc *wc, const struct ib_grh *grh,
3571 struct rdma_ah_attr *ah_attr);
3574 * ib_create_ah_from_wc - Creates an address handle associated with the
3575 * sender of the specified work completion.
3576 * @pd: The protection domain associated with the address handle.
3577 * @wc: Work completion information associated with a received message.
3578 * @grh: References the received global route header. This parameter is
3579 * ignored unless the work completion indicates that the GRH is valid.
3580 * @port_num: The outbound port number to associate with the address.
3582 * The address handle is used to reference a local or global destination
3583 * in all UD QP post sends.
3585 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3586 const struct ib_grh *grh, u32 port_num);
3589 * rdma_modify_ah - Modifies the address vector associated with an address
3591 * @ah: The address handle to modify.
3592 * @ah_attr: The new address vector attributes to associate with the
3595 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3598 * rdma_query_ah - Queries the address vector associated with an address
3600 * @ah: The address handle to query.
3601 * @ah_attr: The address vector attributes associated with the address
3604 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3606 enum rdma_destroy_ah_flags {
3607 /* In a sleepable context */
3608 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3612 * rdma_destroy_ah_user - Destroys an address handle.
3613 * @ah: The address handle to destroy.
3614 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3615 * @udata: Valid user data or NULL for kernel objects
3617 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3620 * rdma_destroy_ah - Destroys an kernel address handle.
3621 * @ah: The address handle to destroy.
3622 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3624 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3626 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3628 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3630 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3633 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3634 struct ib_srq_init_attr *srq_init_attr,
3635 struct ib_usrq_object *uobject,
3636 struct ib_udata *udata);
3637 static inline struct ib_srq *
3638 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3640 if (!pd->device->ops.create_srq)
3641 return ERR_PTR(-EOPNOTSUPP);
3643 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3647 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3648 * @srq: The SRQ to modify.
3649 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3650 * the current values of selected SRQ attributes are returned.
3651 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3652 * are being modified.
3654 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3655 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3656 * the number of receives queued drops below the limit.
3658 int ib_modify_srq(struct ib_srq *srq,
3659 struct ib_srq_attr *srq_attr,
3660 enum ib_srq_attr_mask srq_attr_mask);
3663 * ib_query_srq - Returns the attribute list and current values for the
3665 * @srq: The SRQ to query.
3666 * @srq_attr: The attributes of the specified SRQ.
3668 int ib_query_srq(struct ib_srq *srq,
3669 struct ib_srq_attr *srq_attr);
3672 * ib_destroy_srq_user - Destroys the specified SRQ.
3673 * @srq: The SRQ to destroy.
3674 * @udata: Valid user data or NULL for kernel objects
3676 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3679 * ib_destroy_srq - Destroys the specified kernel SRQ.
3680 * @srq: The SRQ to destroy.
3682 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3684 static inline void ib_destroy_srq(struct ib_srq *srq)
3686 int ret = ib_destroy_srq_user(srq, NULL);
3688 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3692 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3693 * @srq: The SRQ to post the work request on.
3694 * @recv_wr: A list of work requests to post on the receive queue.
3695 * @bad_recv_wr: On an immediate failure, this parameter will reference
3696 * the work request that failed to be posted on the QP.
3698 static inline int ib_post_srq_recv(struct ib_srq *srq,
3699 const struct ib_recv_wr *recv_wr,
3700 const struct ib_recv_wr **bad_recv_wr)
3702 const struct ib_recv_wr *dummy;
3704 return srq->device->ops.post_srq_recv(srq, recv_wr,
3705 bad_recv_wr ? : &dummy);
3708 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3709 struct ib_qp_init_attr *qp_init_attr,
3710 const char *caller);
3712 * ib_create_qp - Creates a kernel QP associated with the specific protection
3714 * @pd: The protection domain associated with the QP.
3715 * @init_attr: A list of initial attributes required to create the
3716 * QP. If QP creation succeeds, then the attributes are updated to
3717 * the actual capabilities of the created QP.
3719 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3720 struct ib_qp_init_attr *init_attr)
3722 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3726 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3727 * @qp: The QP to modify.
3728 * @attr: On input, specifies the QP attributes to modify. On output,
3729 * the current values of selected QP attributes are returned.
3730 * @attr_mask: A bit-mask used to specify which attributes of the QP
3731 * are being modified.
3732 * @udata: pointer to user's input output buffer information
3733 * are being modified.
3734 * It returns 0 on success and returns appropriate error code on error.
3736 int ib_modify_qp_with_udata(struct ib_qp *qp,
3737 struct ib_qp_attr *attr,
3739 struct ib_udata *udata);
3742 * ib_modify_qp - Modifies the attributes for the specified QP and then
3743 * transitions the QP to the given state.
3744 * @qp: The QP to modify.
3745 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3746 * the current values of selected QP attributes are returned.
3747 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3748 * are being modified.
3750 int ib_modify_qp(struct ib_qp *qp,
3751 struct ib_qp_attr *qp_attr,
3755 * ib_query_qp - Returns the attribute list and current values for the
3757 * @qp: The QP to query.
3758 * @qp_attr: The attributes of the specified QP.
3759 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3760 * @qp_init_attr: Additional attributes of the selected QP.
3762 * The qp_attr_mask may be used to limit the query to gathering only the
3763 * selected attributes.
3765 int ib_query_qp(struct ib_qp *qp,
3766 struct ib_qp_attr *qp_attr,
3768 struct ib_qp_init_attr *qp_init_attr);
3771 * ib_destroy_qp - Destroys the specified QP.
3772 * @qp: The QP to destroy.
3773 * @udata: Valid udata or NULL for kernel objects
3775 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3778 * ib_destroy_qp - Destroys the specified kernel QP.
3779 * @qp: The QP to destroy.
3781 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3783 static inline int ib_destroy_qp(struct ib_qp *qp)
3785 return ib_destroy_qp_user(qp, NULL);
3789 * ib_open_qp - Obtain a reference to an existing sharable QP.
3790 * @xrcd - XRC domain
3791 * @qp_open_attr: Attributes identifying the QP to open.
3793 * Returns a reference to a sharable QP.
3795 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3796 struct ib_qp_open_attr *qp_open_attr);
3799 * ib_close_qp - Release an external reference to a QP.
3800 * @qp: The QP handle to release
3802 * The opened QP handle is released by the caller. The underlying
3803 * shared QP is not destroyed until all internal references are released.
3805 int ib_close_qp(struct ib_qp *qp);
3808 * ib_post_send - Posts a list of work requests to the send queue of
3810 * @qp: The QP to post the work request on.
3811 * @send_wr: A list of work requests to post on the send queue.
3812 * @bad_send_wr: On an immediate failure, this parameter will reference
3813 * the work request that failed to be posted on the QP.
3815 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3816 * error is returned, the QP state shall not be affected,
3817 * ib_post_send() will return an immediate error after queueing any
3818 * earlier work requests in the list.
3820 static inline int ib_post_send(struct ib_qp *qp,
3821 const struct ib_send_wr *send_wr,
3822 const struct ib_send_wr **bad_send_wr)
3824 const struct ib_send_wr *dummy;
3826 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3830 * ib_post_recv - Posts a list of work requests to the receive queue of
3832 * @qp: The QP to post the work request on.
3833 * @recv_wr: A list of work requests to post on the receive queue.
3834 * @bad_recv_wr: On an immediate failure, this parameter will reference
3835 * the work request that failed to be posted on the QP.
3837 static inline int ib_post_recv(struct ib_qp *qp,
3838 const struct ib_recv_wr *recv_wr,
3839 const struct ib_recv_wr **bad_recv_wr)
3841 const struct ib_recv_wr *dummy;
3843 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3846 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3847 int comp_vector, enum ib_poll_context poll_ctx,
3848 const char *caller);
3849 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3850 int nr_cqe, int comp_vector,
3851 enum ib_poll_context poll_ctx)
3853 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3857 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3858 int nr_cqe, enum ib_poll_context poll_ctx,
3859 const char *caller);
3862 * ib_alloc_cq_any: Allocate kernel CQ
3863 * @dev: The IB device
3864 * @private: Private data attached to the CQE
3865 * @nr_cqe: Number of CQEs in the CQ
3866 * @poll_ctx: Context used for polling the CQ
3868 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3869 void *private, int nr_cqe,
3870 enum ib_poll_context poll_ctx)
3872 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3876 void ib_free_cq(struct ib_cq *cq);
3877 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3880 * ib_create_cq - Creates a CQ on the specified device.
3881 * @device: The device on which to create the CQ.
3882 * @comp_handler: A user-specified callback that is invoked when a
3883 * completion event occurs on the CQ.
3884 * @event_handler: A user-specified callback that is invoked when an
3885 * asynchronous event not associated with a completion occurs on the CQ.
3886 * @cq_context: Context associated with the CQ returned to the user via
3887 * the associated completion and event handlers.
3888 * @cq_attr: The attributes the CQ should be created upon.
3890 * Users can examine the cq structure to determine the actual CQ size.
3892 struct ib_cq *__ib_create_cq(struct ib_device *device,
3893 ib_comp_handler comp_handler,
3894 void (*event_handler)(struct ib_event *, void *),
3896 const struct ib_cq_init_attr *cq_attr,
3897 const char *caller);
3898 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3899 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3902 * ib_resize_cq - Modifies the capacity of the CQ.
3903 * @cq: The CQ to resize.
3904 * @cqe: The minimum size of the CQ.
3906 * Users can examine the cq structure to determine the actual CQ size.
3908 int ib_resize_cq(struct ib_cq *cq, int cqe);
3911 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3912 * @cq: The CQ to modify.
3913 * @cq_count: number of CQEs that will trigger an event
3914 * @cq_period: max period of time in usec before triggering an event
3917 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3920 * ib_destroy_cq_user - Destroys the specified CQ.
3921 * @cq: The CQ to destroy.
3922 * @udata: Valid user data or NULL for kernel objects
3924 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3927 * ib_destroy_cq - Destroys the specified kernel CQ.
3928 * @cq: The CQ to destroy.
3930 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3932 static inline void ib_destroy_cq(struct ib_cq *cq)
3934 int ret = ib_destroy_cq_user(cq, NULL);
3936 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3940 * ib_poll_cq - poll a CQ for completion(s)
3941 * @cq:the CQ being polled
3942 * @num_entries:maximum number of completions to return
3943 * @wc:array of at least @num_entries &struct ib_wc where completions
3946 * Poll a CQ for (possibly multiple) completions. If the return value
3947 * is < 0, an error occurred. If the return value is >= 0, it is the
3948 * number of completions returned. If the return value is
3949 * non-negative and < num_entries, then the CQ was emptied.
3951 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3954 return cq->device->ops.poll_cq(cq, num_entries, wc);
3958 * ib_req_notify_cq - Request completion notification on a CQ.
3959 * @cq: The CQ to generate an event for.
3961 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3962 * to request an event on the next solicited event or next work
3963 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3964 * may also be |ed in to request a hint about missed events, as
3968 * < 0 means an error occurred while requesting notification
3969 * == 0 means notification was requested successfully, and if
3970 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3971 * were missed and it is safe to wait for another event. In
3972 * this case is it guaranteed that any work completions added
3973 * to the CQ since the last CQ poll will trigger a completion
3974 * notification event.
3975 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3976 * in. It means that the consumer must poll the CQ again to
3977 * make sure it is empty to avoid missing an event because of a
3978 * race between requesting notification and an entry being
3979 * added to the CQ. This return value means it is possible
3980 * (but not guaranteed) that a work completion has been added
3981 * to the CQ since the last poll without triggering a
3982 * completion notification event.
3984 static inline int ib_req_notify_cq(struct ib_cq *cq,
3985 enum ib_cq_notify_flags flags)
3987 return cq->device->ops.req_notify_cq(cq, flags);
3990 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3991 int comp_vector_hint,
3992 enum ib_poll_context poll_ctx);
3994 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3997 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3998 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3999 * address into the dma address.
4001 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4003 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4007 * ib_dma_mapping_error - check a DMA addr for error
4008 * @dev: The device for which the dma_addr was created
4009 * @dma_addr: The DMA address to check
4011 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4013 if (ib_uses_virt_dma(dev))
4015 return dma_mapping_error(dev->dma_device, dma_addr);
4019 * ib_dma_map_single - Map a kernel virtual address to DMA address
4020 * @dev: The device for which the dma_addr is to be created
4021 * @cpu_addr: The kernel virtual address
4022 * @size: The size of the region in bytes
4023 * @direction: The direction of the DMA
4025 static inline u64 ib_dma_map_single(struct ib_device *dev,
4026 void *cpu_addr, size_t size,
4027 enum dma_data_direction direction)
4029 if (ib_uses_virt_dma(dev))
4030 return (uintptr_t)cpu_addr;
4031 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4035 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4036 * @dev: The device for which the DMA address was created
4037 * @addr: The DMA address
4038 * @size: The size of the region in bytes
4039 * @direction: The direction of the DMA
4041 static inline void ib_dma_unmap_single(struct ib_device *dev,
4042 u64 addr, size_t size,
4043 enum dma_data_direction direction)
4045 if (!ib_uses_virt_dma(dev))
4046 dma_unmap_single(dev->dma_device, addr, size, direction);
4050 * ib_dma_map_page - Map a physical page to DMA address
4051 * @dev: The device for which the dma_addr is to be created
4052 * @page: The page to be mapped
4053 * @offset: The offset within the page
4054 * @size: The size of the region in bytes
4055 * @direction: The direction of the DMA
4057 static inline u64 ib_dma_map_page(struct ib_device *dev,
4059 unsigned long offset,
4061 enum dma_data_direction direction)
4063 if (ib_uses_virt_dma(dev))
4064 return (uintptr_t)(page_address(page) + offset);
4065 return dma_map_page(dev->dma_device, page, offset, size, direction);
4069 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4070 * @dev: The device for which the DMA address was created
4071 * @addr: The DMA address
4072 * @size: The size of the region in bytes
4073 * @direction: The direction of the DMA
4075 static inline void ib_dma_unmap_page(struct ib_device *dev,
4076 u64 addr, size_t size,
4077 enum dma_data_direction direction)
4079 if (!ib_uses_virt_dma(dev))
4080 dma_unmap_page(dev->dma_device, addr, size, direction);
4083 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4084 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4085 struct scatterlist *sg, int nents,
4086 enum dma_data_direction direction,
4087 unsigned long dma_attrs)
4089 if (ib_uses_virt_dma(dev))
4090 return ib_dma_virt_map_sg(dev, sg, nents);
4091 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4095 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4096 struct scatterlist *sg, int nents,
4097 enum dma_data_direction direction,
4098 unsigned long dma_attrs)
4100 if (!ib_uses_virt_dma(dev))
4101 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4106 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4107 * @dev: The device for which the DMA addresses are to be created
4108 * @sg: The sg_table object describing the buffer
4109 * @direction: The direction of the DMA
4110 * @attrs: Optional DMA attributes for the map operation
4112 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4113 struct sg_table *sgt,
4114 enum dma_data_direction direction,
4115 unsigned long dma_attrs)
4119 if (ib_uses_virt_dma(dev)) {
4120 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4126 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4129 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4130 struct sg_table *sgt,
4131 enum dma_data_direction direction,
4132 unsigned long dma_attrs)
4134 if (!ib_uses_virt_dma(dev))
4135 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4139 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4140 * @dev: The device for which the DMA addresses are to be created
4141 * @sg: The array of scatter/gather entries
4142 * @nents: The number of scatter/gather entries
4143 * @direction: The direction of the DMA
4145 static inline int ib_dma_map_sg(struct ib_device *dev,
4146 struct scatterlist *sg, int nents,
4147 enum dma_data_direction direction)
4149 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4153 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4154 * @dev: The device for which the DMA addresses were created
4155 * @sg: The array of scatter/gather entries
4156 * @nents: The number of scatter/gather entries
4157 * @direction: The direction of the DMA
4159 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4160 struct scatterlist *sg, int nents,
4161 enum dma_data_direction direction)
4163 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4167 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4168 * @dev: The device to query
4170 * The returned value represents a size in bytes.
4172 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4174 if (ib_uses_virt_dma(dev))
4176 return dma_get_max_seg_size(dev->dma_device);
4180 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4181 * @dev: The device for which the DMA address was created
4182 * @addr: The DMA address
4183 * @size: The size of the region in bytes
4184 * @dir: The direction of the DMA
4186 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4189 enum dma_data_direction dir)
4191 if (!ib_uses_virt_dma(dev))
4192 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4196 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4197 * @dev: The device for which the DMA address was created
4198 * @addr: The DMA address
4199 * @size: The size of the region in bytes
4200 * @dir: The direction of the DMA
4202 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4205 enum dma_data_direction dir)
4207 if (!ib_uses_virt_dma(dev))
4208 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4211 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4212 * space. This function should be called when 'current' is the owning MM.
4214 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4215 u64 virt_addr, int mr_access_flags);
4217 /* ib_advise_mr - give an advice about an address range in a memory region */
4218 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4219 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4221 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4222 * HCA translation table.
4223 * @mr: The memory region to deregister.
4224 * @udata: Valid user data or NULL for kernel object
4226 * This function can fail, if the memory region has memory windows bound to it.
4228 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4231 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4232 * HCA translation table.
4233 * @mr: The memory region to deregister.
4235 * This function can fail, if the memory region has memory windows bound to it.
4237 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4239 static inline int ib_dereg_mr(struct ib_mr *mr)
4241 return ib_dereg_mr_user(mr, NULL);
4244 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4247 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4248 u32 max_num_data_sg,
4249 u32 max_num_meta_sg);
4252 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4254 * @mr - struct ib_mr pointer to be updated.
4255 * @newkey - new key to be used.
4257 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4259 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4260 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4264 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4265 * for calculating a new rkey for type 2 memory windows.
4266 * @rkey - the rkey to increment.
4268 static inline u32 ib_inc_rkey(u32 rkey)
4270 const u32 mask = 0x000000ff;
4271 return ((rkey + 1) & mask) | (rkey & ~mask);
4275 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4276 * @qp: QP to attach to the multicast group. The QP must be type
4278 * @gid: Multicast group GID.
4279 * @lid: Multicast group LID in host byte order.
4281 * In order to send and receive multicast packets, subnet
4282 * administration must have created the multicast group and configured
4283 * the fabric appropriately. The port associated with the specified
4284 * QP must also be a member of the multicast group.
4286 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4289 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4290 * @qp: QP to detach from the multicast group.
4291 * @gid: Multicast group GID.
4292 * @lid: Multicast group LID in host byte order.
4294 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4296 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4297 struct inode *inode, struct ib_udata *udata);
4298 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4300 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4304 * Local write permission is required if remote write or
4305 * remote atomic permission is also requested.
4307 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4308 !(flags & IB_ACCESS_LOCAL_WRITE))
4311 if (flags & ~IB_ACCESS_SUPPORTED)
4314 if (flags & IB_ACCESS_ON_DEMAND &&
4315 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4320 static inline bool ib_access_writable(int access_flags)
4323 * We have writable memory backing the MR if any of the following
4324 * access flags are set. "Local write" and "remote write" obviously
4325 * require write access. "Remote atomic" can do things like fetch and
4326 * add, which will modify memory, and "MW bind" can change permissions
4327 * by binding a window.
4329 return access_flags &
4330 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4331 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4335 * ib_check_mr_status: lightweight check of MR status.
4336 * This routine may provide status checks on a selected
4337 * ib_mr. first use is for signature status check.
4339 * @mr: A memory region.
4340 * @check_mask: Bitmask of which checks to perform from
4341 * ib_mr_status_check enumeration.
4342 * @mr_status: The container of relevant status checks.
4343 * failed checks will be indicated in the status bitmask
4344 * and the relevant info shall be in the error item.
4346 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4347 struct ib_mr_status *mr_status);
4350 * ib_device_try_get: Hold a registration lock
4351 * device: The device to lock
4353 * A device under an active registration lock cannot become unregistered. It
4354 * is only possible to obtain a registration lock on a device that is fully
4355 * registered, otherwise this function returns false.
4357 * The registration lock is only necessary for actions which require the
4358 * device to still be registered. Uses that only require the device pointer to
4359 * be valid should use get_device(&ibdev->dev) to hold the memory.
4362 static inline bool ib_device_try_get(struct ib_device *dev)
4364 return refcount_inc_not_zero(&dev->refcount);
4367 void ib_device_put(struct ib_device *device);
4368 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4369 enum rdma_driver_id driver_id);
4370 struct ib_device *ib_device_get_by_name(const char *name,
4371 enum rdma_driver_id driver_id);
4372 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4373 u16 pkey, const union ib_gid *gid,
4374 const struct sockaddr *addr);
4375 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4377 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4379 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4380 struct ib_wq_init_attr *init_attr);
4381 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4383 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4384 unsigned int *sg_offset, unsigned int page_size);
4385 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4386 int data_sg_nents, unsigned int *data_sg_offset,
4387 struct scatterlist *meta_sg, int meta_sg_nents,
4388 unsigned int *meta_sg_offset, unsigned int page_size);
4391 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4392 unsigned int *sg_offset, unsigned int page_size)
4396 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4402 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4403 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4405 void ib_drain_rq(struct ib_qp *qp);
4406 void ib_drain_sq(struct ib_qp *qp);
4407 void ib_drain_qp(struct ib_qp *qp);
4409 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4412 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4414 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4415 return attr->roce.dmac;
4419 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4421 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4422 attr->ib.dlid = (u16)dlid;
4423 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4424 attr->opa.dlid = dlid;
4427 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4429 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4430 return attr->ib.dlid;
4431 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4432 return attr->opa.dlid;
4436 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4441 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4446 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4449 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4450 attr->ib.src_path_bits = src_path_bits;
4451 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4452 attr->opa.src_path_bits = src_path_bits;
4455 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4457 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4458 return attr->ib.src_path_bits;
4459 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4460 return attr->opa.src_path_bits;
4464 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4467 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4468 attr->opa.make_grd = make_grd;
4471 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4473 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4474 return attr->opa.make_grd;
4478 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4480 attr->port_num = port_num;
4483 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4485 return attr->port_num;
4488 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4491 attr->static_rate = static_rate;
4494 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4496 return attr->static_rate;
4499 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4500 enum ib_ah_flags flag)
4502 attr->ah_flags = flag;
4505 static inline enum ib_ah_flags
4506 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4508 return attr->ah_flags;
4511 static inline const struct ib_global_route
4512 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4517 /*To retrieve and modify the grh */
4518 static inline struct ib_global_route
4519 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4524 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4526 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4528 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4531 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4534 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4536 grh->dgid.global.subnet_prefix = prefix;
4539 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4542 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4544 grh->dgid.global.interface_id = if_id;
4547 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4548 union ib_gid *dgid, u32 flow_label,
4549 u8 sgid_index, u8 hop_limit,
4552 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4554 attr->ah_flags = IB_AH_GRH;
4557 grh->flow_label = flow_label;
4558 grh->sgid_index = sgid_index;
4559 grh->hop_limit = hop_limit;
4560 grh->traffic_class = traffic_class;
4561 grh->sgid_attr = NULL;
4564 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4565 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4566 u32 flow_label, u8 hop_limit, u8 traffic_class,
4567 const struct ib_gid_attr *sgid_attr);
4568 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4569 const struct rdma_ah_attr *src);
4570 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4571 const struct rdma_ah_attr *new);
4572 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4575 * rdma_ah_find_type - Return address handle type.
4577 * @dev: Device to be checked
4578 * @port_num: Port number
4580 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4583 if (rdma_protocol_roce(dev, port_num))
4584 return RDMA_AH_ATTR_TYPE_ROCE;
4585 if (rdma_protocol_ib(dev, port_num)) {
4586 if (rdma_cap_opa_ah(dev, port_num))
4587 return RDMA_AH_ATTR_TYPE_OPA;
4588 return RDMA_AH_ATTR_TYPE_IB;
4591 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4595 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4596 * In the current implementation the only way to get
4597 * get the 32bit lid is from other sources for OPA.
4598 * For IB, lids will always be 16bits so cast the
4599 * value accordingly.
4603 static inline u16 ib_lid_cpu16(u32 lid)
4605 WARN_ON_ONCE(lid & 0xFFFF0000);
4610 * ib_lid_be16 - Return lid in 16bit BE encoding.
4614 static inline __be16 ib_lid_be16(u32 lid)
4616 WARN_ON_ONCE(lid & 0xFFFF0000);
4617 return cpu_to_be16((u16)lid);
4621 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4623 * @device: the rdma device
4624 * @comp_vector: index of completion vector
4626 * Returns NULL on failure, otherwise a corresponding cpu map of the
4627 * completion vector (returns all-cpus map if the device driver doesn't
4628 * implement get_vector_affinity).
4630 static inline const struct cpumask *
4631 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4633 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4634 !device->ops.get_vector_affinity)
4637 return device->ops.get_vector_affinity(device, comp_vector);
4642 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4643 * and add their gids, as needed, to the relevant RoCE devices.
4645 * @device: the rdma device
4647 void rdma_roce_rescan_device(struct ib_device *ibdev);
4649 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4651 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4653 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4654 enum rdma_netdev_t type, const char *name,
4655 unsigned char name_assign_type,
4656 void (*setup)(struct net_device *));
4658 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4659 enum rdma_netdev_t type, const char *name,
4660 unsigned char name_assign_type,
4661 void (*setup)(struct net_device *),
4662 struct net_device *netdev);
4665 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4667 * @device: device pointer for which ib_device pointer to retrieve
4669 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4672 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4674 struct ib_core_device *coredev =
4675 container_of(device, struct ib_core_device, dev);
4677 return coredev->owner;
4681 * ibdev_to_node - return the NUMA node for a given ib_device
4682 * @dev: device to get the NUMA node for.
4684 static inline int ibdev_to_node(struct ib_device *ibdev)
4686 struct device *parent = ibdev->dev.parent;
4689 return NUMA_NO_NODE;
4690 return dev_to_node(parent);
4694 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4695 * ib_device holder structure from device pointer.
4697 * NOTE: New drivers should not make use of this API; This API is only for
4698 * existing drivers who have exposed sysfs entries using
4699 * ops->device_group.
4701 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4702 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4704 bool rdma_dev_access_netns(const struct ib_device *device,
4705 const struct net *net);
4707 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4708 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4709 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4712 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4715 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4716 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4719 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4721 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4723 fl_low ^= fl_high >> 14;
4724 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4728 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4729 * local and remote qpn values
4731 * This function folded the multiplication results of two qpns, 24 bit each,
4732 * fields, and converts it to a 20 bit results.
4734 * This function will create symmetric flow_label value based on the local
4735 * and remote qpn values. this will allow both the requester and responder
4736 * to calculate the same flow_label for a given connection.
4738 * This helper function should be used by driver in case the upper layer
4739 * provide a zero flow_label value. This is to improve entropy of RDMA
4740 * traffic in the network.
4742 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4744 u64 v = (u64)lqpn * rqpn;
4749 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4752 const struct ib_port_immutable*
4753 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4754 #endif /* IB_VERBS_H */