2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
72 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
76 extern struct workqueue_struct *ib_wq;
77 extern struct workqueue_struct *ib_comp_wq;
78 extern struct workqueue_struct *ib_comp_unbound_wq;
81 void ibdev_printk(const char *level, const struct ib_device *ibdev,
82 const char *format, ...);
84 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
86 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
88 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
90 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
92 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
94 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
96 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
98 #if defined(CONFIG_DYNAMIC_DEBUG)
99 #define ibdev_dbg(__dev, format, args...) \
100 dynamic_ibdev_dbg(__dev, format, ##args)
102 __printf(2, 3) __cold
104 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
107 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
109 static DEFINE_RATELIMIT_STATE(_rs, \
110 DEFAULT_RATELIMIT_INTERVAL, \
111 DEFAULT_RATELIMIT_BURST); \
112 if (__ratelimit(&_rs)) \
113 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
116 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
117 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
118 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
119 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
120 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
121 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
122 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
123 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
124 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
125 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
126 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
127 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
128 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
129 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
131 #if defined(CONFIG_DYNAMIC_DEBUG)
132 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
133 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
135 static DEFINE_RATELIMIT_STATE(_rs, \
136 DEFAULT_RATELIMIT_INTERVAL, \
137 DEFAULT_RATELIMIT_BURST); \
138 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
139 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
140 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
144 __printf(2, 3) __cold
146 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
152 __be64 subnet_prefix;
157 extern union ib_gid zgid;
160 /* If link layer is Ethernet, this is RoCE V1 */
162 IB_GID_TYPE_ROCE = 0,
163 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
167 #define ROCE_V2_UDP_DPORT 4791
169 struct net_device __rcu *ndev;
170 struct ib_device *device;
172 enum ib_gid_type gid_type;
178 /* set the local administered indication */
179 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
182 enum rdma_transport_type {
184 RDMA_TRANSPORT_IWARP,
185 RDMA_TRANSPORT_USNIC,
186 RDMA_TRANSPORT_USNIC_UDP,
187 RDMA_TRANSPORT_UNSPECIFIED,
190 enum rdma_protocol_type {
194 RDMA_PROTOCOL_USNIC_UDP
197 __attribute_const__ enum rdma_transport_type
198 rdma_node_get_transport(unsigned int node_type);
200 enum rdma_network_type {
202 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
207 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
209 if (network_type == RDMA_NETWORK_IPV4 ||
210 network_type == RDMA_NETWORK_IPV6)
211 return IB_GID_TYPE_ROCE_UDP_ENCAP;
213 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
214 return IB_GID_TYPE_IB;
217 static inline enum rdma_network_type
218 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
220 if (attr->gid_type == IB_GID_TYPE_IB)
221 return RDMA_NETWORK_IB;
223 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
224 return RDMA_NETWORK_IPV4;
226 return RDMA_NETWORK_IPV6;
229 enum rdma_link_layer {
230 IB_LINK_LAYER_UNSPECIFIED,
231 IB_LINK_LAYER_INFINIBAND,
232 IB_LINK_LAYER_ETHERNET,
235 enum ib_device_cap_flags {
236 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
237 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
238 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
239 IB_DEVICE_RAW_MULTI = (1 << 3),
240 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
241 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
242 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
243 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
244 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
245 /* Not in use, former INIT_TYPE = (1 << 9),*/
246 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
247 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
248 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
249 IB_DEVICE_SRQ_RESIZE = (1 << 13),
250 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
253 * This device supports a per-device lkey or stag that can be
254 * used without performing a memory registration for the local
255 * memory. Note that ULPs should never check this flag, but
256 * instead of use the local_dma_lkey flag in the ib_pd structure,
257 * which will always contain a usable lkey.
259 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
260 /* Reserved, old SEND_W_INV = (1 << 16),*/
261 IB_DEVICE_MEM_WINDOW = (1 << 17),
263 * Devices should set IB_DEVICE_UD_IP_SUM if they support
264 * insertion of UDP and TCP checksum on outgoing UD IPoIB
265 * messages and can verify the validity of checksum for
266 * incoming messages. Setting this flag implies that the
267 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
269 IB_DEVICE_UD_IP_CSUM = (1 << 18),
270 IB_DEVICE_UD_TSO = (1 << 19),
271 IB_DEVICE_XRC = (1 << 20),
274 * This device supports the IB "base memory management extension",
275 * which includes support for fast registrations (IB_WR_REG_MR,
276 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
277 * also be set by any iWarp device which must support FRs to comply
278 * to the iWarp verbs spec. iWarp devices also support the
279 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
282 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
283 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
284 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
285 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
286 IB_DEVICE_RC_IP_CSUM = (1 << 25),
287 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
288 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
290 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
291 * support execution of WQEs that involve synchronization
292 * of I/O operations with single completion queue managed
295 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
296 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
297 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
298 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
299 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
300 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
301 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
302 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
303 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
304 /* The device supports padding incoming writes to cacheline. */
305 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
306 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
315 enum ib_odp_general_cap_bits {
316 IB_ODP_SUPPORT = 1 << 0,
317 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
320 enum ib_odp_transport_cap_bits {
321 IB_ODP_SUPPORT_SEND = 1 << 0,
322 IB_ODP_SUPPORT_RECV = 1 << 1,
323 IB_ODP_SUPPORT_WRITE = 1 << 2,
324 IB_ODP_SUPPORT_READ = 1 << 3,
325 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
326 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
330 uint64_t general_caps;
332 uint32_t rc_odp_caps;
333 uint32_t uc_odp_caps;
334 uint32_t ud_odp_caps;
335 uint32_t xrc_odp_caps;
336 } per_transport_caps;
340 /* Corresponding bit will be set if qp type from
341 * 'enum ib_qp_type' is supported, e.g.
342 * supported_qpts |= 1 << IB_QPT_UD
345 u32 max_rwq_indirection_tables;
346 u32 max_rwq_indirection_table_size;
349 enum ib_tm_cap_flags {
350 /* Support tag matching with rendezvous offload for RC transport */
351 IB_TM_CAP_RNDV_RC = 1 << 0,
355 /* Max size of RNDV header */
356 u32 max_rndv_hdr_size;
357 /* Max number of entries in tag matching list */
359 /* From enum ib_tm_cap_flags */
361 /* Max number of outstanding list operations */
363 /* Max number of SGE in tag matching entry */
367 struct ib_cq_init_attr {
373 enum ib_cq_attr_mask {
374 IB_CQ_MODERATE = 1 << 0,
378 u16 max_cq_moderation_count;
379 u16 max_cq_moderation_period;
382 struct ib_dm_mr_attr {
388 struct ib_dm_alloc_attr {
394 struct ib_device_attr {
396 __be64 sys_image_guid;
404 u64 device_cap_flags;
415 int max_qp_init_rd_atom;
416 int max_ee_init_rd_atom;
417 enum ib_atomic_cap atomic_cap;
418 enum ib_atomic_cap masked_atomic_cap;
425 int max_mcast_qp_attach;
426 int max_total_mcast_qp_attach;
433 unsigned int max_fast_reg_page_list_len;
434 unsigned int max_pi_fast_reg_page_list_len;
436 u8 local_ca_ack_delay;
439 struct ib_odp_caps odp_caps;
440 uint64_t timestamp_mask;
441 uint64_t hca_core_clock; /* in KHZ */
442 struct ib_rss_caps rss_caps;
444 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
445 struct ib_tm_caps tm_caps;
446 struct ib_cq_caps cq_caps;
448 /* Max entries for sgl for optimized performance per READ */
460 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
463 case IB_MTU_256: return 256;
464 case IB_MTU_512: return 512;
465 case IB_MTU_1024: return 1024;
466 case IB_MTU_2048: return 2048;
467 case IB_MTU_4096: return 4096;
472 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
476 else if (mtu >= 2048)
478 else if (mtu >= 1024)
492 IB_PORT_ACTIVE_DEFER = 5
495 enum ib_port_phys_state {
496 IB_PORT_PHYS_STATE_SLEEP = 1,
497 IB_PORT_PHYS_STATE_POLLING = 2,
498 IB_PORT_PHYS_STATE_DISABLED = 3,
499 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
500 IB_PORT_PHYS_STATE_LINK_UP = 5,
501 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
502 IB_PORT_PHYS_STATE_PHY_TEST = 7,
513 static inline int ib_width_enum_to_int(enum ib_port_width width)
516 case IB_WIDTH_1X: return 1;
517 case IB_WIDTH_2X: return 2;
518 case IB_WIDTH_4X: return 4;
519 case IB_WIDTH_8X: return 8;
520 case IB_WIDTH_12X: return 12;
536 * struct rdma_hw_stats
537 * @lock - Mutex to protect parallel write access to lifespan and values
538 * of counters, which are 64bits and not guaranteeed to be written
539 * atomicaly on 32bits systems.
540 * @timestamp - Used by the core code to track when the last update was
541 * @lifespan - Used by the core code to determine how old the counters
542 * should be before being updated again. Stored in jiffies, defaults
543 * to 10 milliseconds, drivers can override the default be specifying
544 * their own value during their allocation routine.
545 * @name - Array of pointers to static names used for the counters in
547 * @num_counters - How many hardware counters there are. If name is
548 * shorter than this number, a kernel oops will result. Driver authors
549 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
550 * in their code to prevent this.
551 * @value - Array of u64 counters that are accessed by the sysfs code and
552 * filled in by the drivers get_stats routine
554 struct rdma_hw_stats {
555 struct mutex lock; /* Protect lifespan and values[] */
556 unsigned long timestamp;
557 unsigned long lifespan;
558 const char * const *names;
563 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
565 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
567 * @names - Array of static const char *
568 * @num_counters - How many elements in array
569 * @lifespan - How many milliseconds between updates
571 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
572 const char * const *names, int num_counters,
573 unsigned long lifespan)
575 struct rdma_hw_stats *stats;
577 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
581 stats->names = names;
582 stats->num_counters = num_counters;
583 stats->lifespan = msecs_to_jiffies(lifespan);
589 /* Define bits for the various functionality this port needs to be supported by
592 /* Management 0x00000FFF */
593 #define RDMA_CORE_CAP_IB_MAD 0x00000001
594 #define RDMA_CORE_CAP_IB_SMI 0x00000002
595 #define RDMA_CORE_CAP_IB_CM 0x00000004
596 #define RDMA_CORE_CAP_IW_CM 0x00000008
597 #define RDMA_CORE_CAP_IB_SA 0x00000010
598 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
600 /* Address format 0x000FF000 */
601 #define RDMA_CORE_CAP_AF_IB 0x00001000
602 #define RDMA_CORE_CAP_ETH_AH 0x00002000
603 #define RDMA_CORE_CAP_OPA_AH 0x00004000
604 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
606 /* Protocol 0xFFF00000 */
607 #define RDMA_CORE_CAP_PROT_IB 0x00100000
608 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
609 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
610 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
611 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
612 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
614 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
615 | RDMA_CORE_CAP_PROT_ROCE \
616 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
618 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
619 | RDMA_CORE_CAP_IB_MAD \
620 | RDMA_CORE_CAP_IB_SMI \
621 | RDMA_CORE_CAP_IB_CM \
622 | RDMA_CORE_CAP_IB_SA \
623 | RDMA_CORE_CAP_AF_IB)
624 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
625 | RDMA_CORE_CAP_IB_MAD \
626 | RDMA_CORE_CAP_IB_CM \
627 | RDMA_CORE_CAP_AF_IB \
628 | RDMA_CORE_CAP_ETH_AH)
629 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
630 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
631 | RDMA_CORE_CAP_IB_MAD \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_AF_IB \
634 | RDMA_CORE_CAP_ETH_AH)
635 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
636 | RDMA_CORE_CAP_IW_CM)
637 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
638 | RDMA_CORE_CAP_OPA_MAD)
640 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
642 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
644 struct ib_port_attr {
646 enum ib_port_state state;
648 enum ib_mtu active_mtu;
650 unsigned int ip_gids:1;
651 /* This is the value from PortInfo CapabilityMask, defined by IBA */
670 enum ib_device_modify_flags {
671 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
672 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
675 #define IB_DEVICE_NODE_DESC_MAX 64
677 struct ib_device_modify {
679 char node_desc[IB_DEVICE_NODE_DESC_MAX];
682 enum ib_port_modify_flags {
683 IB_PORT_SHUTDOWN = 1,
684 IB_PORT_INIT_TYPE = (1<<2),
685 IB_PORT_RESET_QKEY_CNTR = (1<<3),
686 IB_PORT_OPA_MASK_CHG = (1<<4)
689 struct ib_port_modify {
690 u32 set_port_cap_mask;
691 u32 clr_port_cap_mask;
699 IB_EVENT_QP_ACCESS_ERR,
703 IB_EVENT_PATH_MIG_ERR,
704 IB_EVENT_DEVICE_FATAL,
705 IB_EVENT_PORT_ACTIVE,
708 IB_EVENT_PKEY_CHANGE,
711 IB_EVENT_SRQ_LIMIT_REACHED,
712 IB_EVENT_QP_LAST_WQE_REACHED,
713 IB_EVENT_CLIENT_REREGISTER,
718 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
721 struct ib_device *device;
729 enum ib_event_type event;
732 struct ib_event_handler {
733 struct ib_device *device;
734 void (*handler)(struct ib_event_handler *, struct ib_event *);
735 struct list_head list;
738 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
740 (_ptr)->device = _device; \
741 (_ptr)->handler = _handler; \
742 INIT_LIST_HEAD(&(_ptr)->list); \
745 struct ib_global_route {
746 const struct ib_gid_attr *sgid_attr;
755 __be32 version_tclass_flow;
763 union rdma_network_hdr {
766 /* The IB spec states that if it's IPv4, the header
767 * is located in the last 20 bytes of the header.
770 struct iphdr roce4grh;
774 #define IB_QPN_MASK 0xFFFFFF
777 IB_MULTICAST_QPN = 0xffffff
780 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
781 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
788 IB_RATE_PORT_CURRENT = 0,
789 IB_RATE_2_5_GBPS = 2,
797 IB_RATE_120_GBPS = 10,
798 IB_RATE_14_GBPS = 11,
799 IB_RATE_56_GBPS = 12,
800 IB_RATE_112_GBPS = 13,
801 IB_RATE_168_GBPS = 14,
802 IB_RATE_25_GBPS = 15,
803 IB_RATE_100_GBPS = 16,
804 IB_RATE_200_GBPS = 17,
805 IB_RATE_300_GBPS = 18,
806 IB_RATE_28_GBPS = 19,
807 IB_RATE_50_GBPS = 20,
808 IB_RATE_400_GBPS = 21,
809 IB_RATE_600_GBPS = 22,
813 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
814 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
815 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
816 * @rate: rate to convert.
818 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
821 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
822 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
823 * @rate: rate to convert.
825 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
829 * enum ib_mr_type - memory region type
830 * @IB_MR_TYPE_MEM_REG: memory region that is used for
831 * normal registration
832 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
833 * register any arbitrary sg lists (without
834 * the normal mr constraints - see
836 * @IB_MR_TYPE_DM: memory region that is used for device
837 * memory registration
838 * @IB_MR_TYPE_USER: memory region that is used for the user-space
840 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
841 * without address translations (VA=PA)
842 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
843 * data integrity operations
851 IB_MR_TYPE_INTEGRITY,
854 enum ib_mr_status_check {
855 IB_MR_CHECK_SIG_STATUS = 1,
859 * struct ib_mr_status - Memory region status container
861 * @fail_status: Bitmask of MR checks status. For each
862 * failed check a corresponding status bit is set.
863 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
866 struct ib_mr_status {
868 struct ib_sig_err sig_err;
872 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
874 * @mult: multiple to convert.
876 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
878 enum rdma_ah_attr_type {
879 RDMA_AH_ATTR_TYPE_UNDEFINED,
880 RDMA_AH_ATTR_TYPE_IB,
881 RDMA_AH_ATTR_TYPE_ROCE,
882 RDMA_AH_ATTR_TYPE_OPA,
890 struct roce_ah_attr {
900 struct rdma_ah_attr {
901 struct ib_global_route grh;
906 enum rdma_ah_attr_type type;
908 struct ib_ah_attr ib;
909 struct roce_ah_attr roce;
910 struct opa_ah_attr opa;
918 IB_WC_LOC_EEC_OP_ERR,
923 IB_WC_LOC_ACCESS_ERR,
924 IB_WC_REM_INV_REQ_ERR,
925 IB_WC_REM_ACCESS_ERR,
928 IB_WC_RNR_RETRY_EXC_ERR,
929 IB_WC_LOC_RDD_VIOL_ERR,
930 IB_WC_REM_INV_RD_REQ_ERR,
933 IB_WC_INV_EEC_STATE_ERR,
935 IB_WC_RESP_TIMEOUT_ERR,
939 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
950 IB_WC_MASKED_COMP_SWAP,
951 IB_WC_MASKED_FETCH_ADD,
953 * Set value of IB_WC_RECV so consumers can test if a completion is a
954 * receive by testing (opcode & IB_WC_RECV).
957 IB_WC_RECV_RDMA_WITH_IMM
962 IB_WC_WITH_IMM = (1<<1),
963 IB_WC_WITH_INVALIDATE = (1<<2),
964 IB_WC_IP_CSUM_OK = (1<<3),
965 IB_WC_WITH_SMAC = (1<<4),
966 IB_WC_WITH_VLAN = (1<<5),
967 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
973 struct ib_cqe *wr_cqe;
975 enum ib_wc_status status;
976 enum ib_wc_opcode opcode;
990 u8 port_num; /* valid only for DR SMPs on switches */
996 enum ib_cq_notify_flags {
997 IB_CQ_SOLICITED = 1 << 0,
998 IB_CQ_NEXT_COMP = 1 << 1,
999 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1000 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1009 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1011 return srq_type == IB_SRQT_XRC ||
1012 srq_type == IB_SRQT_TM;
1015 enum ib_srq_attr_mask {
1016 IB_SRQ_MAX_WR = 1 << 0,
1017 IB_SRQ_LIMIT = 1 << 1,
1020 struct ib_srq_attr {
1026 struct ib_srq_init_attr {
1027 void (*event_handler)(struct ib_event *, void *);
1029 struct ib_srq_attr attr;
1030 enum ib_srq_type srq_type;
1036 struct ib_xrcd *xrcd;
1051 u32 max_inline_data;
1054 * Maximum number of rdma_rw_ctx structures in flight at a time.
1055 * ib_create_qp() will calculate the right amount of neededed WRs
1056 * and MRs based on this.
1068 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1069 * here (and in that order) since the MAD layer uses them as
1070 * indices into a 2-entry table.
1079 IB_QPT_RAW_ETHERTYPE,
1080 IB_QPT_RAW_PACKET = 8,
1084 IB_QPT_DRIVER = 0xFF,
1085 /* Reserve a range for qp types internal to the low level driver.
1086 * These qp types will not be visible at the IB core layer, so the
1087 * IB_QPT_MAX usages should not be affected in the core layer
1089 IB_QPT_RESERVED1 = 0x1000,
1101 enum ib_qp_create_flags {
1102 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1103 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1104 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1105 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1106 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1107 IB_QP_CREATE_NETIF_QP = 1 << 5,
1108 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1109 /* FREE = 1 << 7, */
1110 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1111 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1112 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1113 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1114 /* reserve bits 26-31 for low level drivers' internal use */
1115 IB_QP_CREATE_RESERVED_START = 1 << 26,
1116 IB_QP_CREATE_RESERVED_END = 1 << 31,
1120 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1121 * callback to destroy the passed in QP.
1124 struct ib_qp_init_attr {
1125 /* Consumer's event_handler callback must not block */
1126 void (*event_handler)(struct ib_event *, void *);
1129 struct ib_cq *send_cq;
1130 struct ib_cq *recv_cq;
1132 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1133 struct ib_qp_cap cap;
1134 enum ib_sig_type sq_sig_type;
1135 enum ib_qp_type qp_type;
1139 * Only needed for special QP types, or when using the RW API.
1142 struct ib_rwq_ind_table *rwq_ind_tbl;
1146 struct ib_qp_open_attr {
1147 void (*event_handler)(struct ib_event *, void *);
1150 enum ib_qp_type qp_type;
1153 enum ib_rnr_timeout {
1154 IB_RNR_TIMER_655_36 = 0,
1155 IB_RNR_TIMER_000_01 = 1,
1156 IB_RNR_TIMER_000_02 = 2,
1157 IB_RNR_TIMER_000_03 = 3,
1158 IB_RNR_TIMER_000_04 = 4,
1159 IB_RNR_TIMER_000_06 = 5,
1160 IB_RNR_TIMER_000_08 = 6,
1161 IB_RNR_TIMER_000_12 = 7,
1162 IB_RNR_TIMER_000_16 = 8,
1163 IB_RNR_TIMER_000_24 = 9,
1164 IB_RNR_TIMER_000_32 = 10,
1165 IB_RNR_TIMER_000_48 = 11,
1166 IB_RNR_TIMER_000_64 = 12,
1167 IB_RNR_TIMER_000_96 = 13,
1168 IB_RNR_TIMER_001_28 = 14,
1169 IB_RNR_TIMER_001_92 = 15,
1170 IB_RNR_TIMER_002_56 = 16,
1171 IB_RNR_TIMER_003_84 = 17,
1172 IB_RNR_TIMER_005_12 = 18,
1173 IB_RNR_TIMER_007_68 = 19,
1174 IB_RNR_TIMER_010_24 = 20,
1175 IB_RNR_TIMER_015_36 = 21,
1176 IB_RNR_TIMER_020_48 = 22,
1177 IB_RNR_TIMER_030_72 = 23,
1178 IB_RNR_TIMER_040_96 = 24,
1179 IB_RNR_TIMER_061_44 = 25,
1180 IB_RNR_TIMER_081_92 = 26,
1181 IB_RNR_TIMER_122_88 = 27,
1182 IB_RNR_TIMER_163_84 = 28,
1183 IB_RNR_TIMER_245_76 = 29,
1184 IB_RNR_TIMER_327_68 = 30,
1185 IB_RNR_TIMER_491_52 = 31
1188 enum ib_qp_attr_mask {
1190 IB_QP_CUR_STATE = (1<<1),
1191 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1192 IB_QP_ACCESS_FLAGS = (1<<3),
1193 IB_QP_PKEY_INDEX = (1<<4),
1194 IB_QP_PORT = (1<<5),
1195 IB_QP_QKEY = (1<<6),
1197 IB_QP_PATH_MTU = (1<<8),
1198 IB_QP_TIMEOUT = (1<<9),
1199 IB_QP_RETRY_CNT = (1<<10),
1200 IB_QP_RNR_RETRY = (1<<11),
1201 IB_QP_RQ_PSN = (1<<12),
1202 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1203 IB_QP_ALT_PATH = (1<<14),
1204 IB_QP_MIN_RNR_TIMER = (1<<15),
1205 IB_QP_SQ_PSN = (1<<16),
1206 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1207 IB_QP_PATH_MIG_STATE = (1<<18),
1208 IB_QP_CAP = (1<<19),
1209 IB_QP_DEST_QPN = (1<<20),
1210 IB_QP_RESERVED1 = (1<<21),
1211 IB_QP_RESERVED2 = (1<<22),
1212 IB_QP_RESERVED3 = (1<<23),
1213 IB_QP_RESERVED4 = (1<<24),
1214 IB_QP_RATE_LIMIT = (1<<25),
1239 enum ib_qp_state qp_state;
1240 enum ib_qp_state cur_qp_state;
1241 enum ib_mtu path_mtu;
1242 enum ib_mig_state path_mig_state;
1247 int qp_access_flags;
1248 struct ib_qp_cap cap;
1249 struct rdma_ah_attr ah_attr;
1250 struct rdma_ah_attr alt_ah_attr;
1253 u8 en_sqd_async_notify;
1256 u8 max_dest_rd_atomic;
1268 /* These are shared with userspace */
1269 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1270 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1271 IB_WR_SEND = IB_UVERBS_WR_SEND,
1272 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1273 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1274 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1275 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1276 IB_WR_LSO = IB_UVERBS_WR_TSO,
1277 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1278 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1279 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1280 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1281 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1282 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1283 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1285 /* These are kernel only and can not be issued by userspace */
1286 IB_WR_REG_MR = 0x20,
1287 IB_WR_REG_MR_INTEGRITY,
1289 /* reserve values for low level drivers' internal use.
1290 * These values will not be used at all in the ib core layer.
1292 IB_WR_RESERVED1 = 0xf0,
1304 enum ib_send_flags {
1306 IB_SEND_SIGNALED = (1<<1),
1307 IB_SEND_SOLICITED = (1<<2),
1308 IB_SEND_INLINE = (1<<3),
1309 IB_SEND_IP_CSUM = (1<<4),
1311 /* reserve bits 26-31 for low level drivers' internal use */
1312 IB_SEND_RESERVED_START = (1 << 26),
1313 IB_SEND_RESERVED_END = (1 << 31),
1323 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1327 struct ib_send_wr *next;
1330 struct ib_cqe *wr_cqe;
1332 struct ib_sge *sg_list;
1334 enum ib_wr_opcode opcode;
1338 u32 invalidate_rkey;
1343 struct ib_send_wr wr;
1348 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1350 return container_of(wr, struct ib_rdma_wr, wr);
1353 struct ib_atomic_wr {
1354 struct ib_send_wr wr;
1358 u64 compare_add_mask;
1363 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1365 return container_of(wr, struct ib_atomic_wr, wr);
1369 struct ib_send_wr wr;
1376 u16 pkey_index; /* valid for GSI only */
1377 u8 port_num; /* valid for DR SMPs on switch only */
1380 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1382 return container_of(wr, struct ib_ud_wr, wr);
1386 struct ib_send_wr wr;
1392 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1394 return container_of(wr, struct ib_reg_wr, wr);
1398 struct ib_recv_wr *next;
1401 struct ib_cqe *wr_cqe;
1403 struct ib_sge *sg_list;
1407 enum ib_access_flags {
1408 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1409 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1410 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1411 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1412 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1413 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1414 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1415 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1417 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1421 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1422 * are hidden here instead of a uapi header!
1424 enum ib_mr_rereg_flags {
1425 IB_MR_REREG_TRANS = 1,
1426 IB_MR_REREG_PD = (1<<1),
1427 IB_MR_REREG_ACCESS = (1<<2),
1428 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1431 struct ib_fmr_attr {
1439 enum rdma_remove_reason {
1441 * Userspace requested uobject deletion or initial try
1442 * to remove uobject via cleanup. Call could fail
1444 RDMA_REMOVE_DESTROY,
1445 /* Context deletion. This call should delete the actual object itself */
1447 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1448 RDMA_REMOVE_DRIVER_REMOVE,
1449 /* uobj is being cleaned-up before being committed */
1453 struct ib_rdmacg_object {
1454 #ifdef CONFIG_CGROUP_RDMA
1455 struct rdma_cgroup *cg; /* owner rdma cgroup */
1459 struct ib_ucontext {
1460 struct ib_device *device;
1461 struct ib_uverbs_file *ufile;
1463 * 'closing' can be read by the driver only during a destroy callback,
1464 * it is set when we are closing the file descriptor and indicates
1465 * that mm_sem may be locked.
1469 bool cleanup_retryable;
1471 struct ib_rdmacg_object cg_obj;
1473 * Implementation details of the RDMA core, don't use in drivers:
1475 struct rdma_restrack_entry res;
1476 struct xarray mmap_xa;
1480 u64 user_handle; /* handle given to us by userspace */
1481 /* ufile & ucontext owning this object */
1482 struct ib_uverbs_file *ufile;
1483 /* FIXME, save memory: ufile->context == context */
1484 struct ib_ucontext *context; /* associated user context */
1485 void *object; /* containing object */
1486 struct list_head list; /* link to context's list */
1487 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1488 int id; /* index into kernel idr */
1490 atomic_t usecnt; /* protects exclusive access */
1491 struct rcu_head rcu; /* kfree_rcu() overhead */
1493 const struct uverbs_api_object *uapi_object;
1497 const void __user *inbuf;
1498 void __user *outbuf;
1506 struct ib_device *device;
1507 struct ib_uobject *uobject;
1508 atomic_t usecnt; /* count all resources */
1510 u32 unsafe_global_rkey;
1513 * Implementation details of the RDMA core, don't use in drivers:
1515 struct ib_mr *__internal_mr;
1516 struct rdma_restrack_entry res;
1520 struct ib_device *device;
1521 atomic_t usecnt; /* count all exposed resources */
1522 struct inode *inode;
1524 struct mutex tgt_qp_mutex;
1525 struct list_head tgt_qp_list;
1529 struct ib_device *device;
1531 struct ib_uobject *uobject;
1532 const struct ib_gid_attr *sgid_attr;
1533 enum rdma_ah_attr_type type;
1536 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1538 enum ib_poll_context {
1539 IB_POLL_DIRECT, /* caller context, no hw completions */
1540 IB_POLL_SOFTIRQ, /* poll from softirq context */
1541 IB_POLL_WORKQUEUE, /* poll from workqueue */
1542 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1546 struct ib_device *device;
1547 struct ib_uobject *uobject;
1548 ib_comp_handler comp_handler;
1549 void (*event_handler)(struct ib_event *, void *);
1552 atomic_t usecnt; /* count number of work queues */
1553 enum ib_poll_context poll_ctx;
1556 struct irq_poll iop;
1557 struct work_struct work;
1559 struct workqueue_struct *comp_wq;
1562 * Implementation details of the RDMA core, don't use in drivers:
1564 struct rdma_restrack_entry res;
1568 struct ib_device *device;
1570 struct ib_uobject *uobject;
1571 void (*event_handler)(struct ib_event *, void *);
1573 enum ib_srq_type srq_type;
1580 struct ib_xrcd *xrcd;
1587 enum ib_raw_packet_caps {
1588 /* Strip cvlan from incoming packet and report it in the matching work
1589 * completion is supported.
1591 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1592 /* Scatter FCS field of an incoming packet to host memory is supported.
1594 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1595 /* Checksum offloads are supported (for both send and receive). */
1596 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1597 /* When a packet is received for an RQ with no receive WQEs, the
1598 * packet processing is delayed.
1600 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1614 struct ib_device *device;
1615 struct ib_uobject *uobject;
1617 void (*event_handler)(struct ib_event *, void *);
1621 enum ib_wq_state state;
1622 enum ib_wq_type wq_type;
1627 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1628 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1629 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1630 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1633 struct ib_wq_init_attr {
1635 enum ib_wq_type wq_type;
1639 void (*event_handler)(struct ib_event *, void *);
1640 u32 create_flags; /* Use enum ib_wq_flags */
1643 enum ib_wq_attr_mask {
1644 IB_WQ_STATE = 1 << 0,
1645 IB_WQ_CUR_STATE = 1 << 1,
1646 IB_WQ_FLAGS = 1 << 2,
1650 enum ib_wq_state wq_state;
1651 enum ib_wq_state curr_wq_state;
1652 u32 flags; /* Use enum ib_wq_flags */
1653 u32 flags_mask; /* Use enum ib_wq_flags */
1656 struct ib_rwq_ind_table {
1657 struct ib_device *device;
1658 struct ib_uobject *uobject;
1661 u32 log_ind_tbl_size;
1662 struct ib_wq **ind_tbl;
1665 struct ib_rwq_ind_table_init_attr {
1666 u32 log_ind_tbl_size;
1667 /* Each entry is a pointer to Receive Work Queue */
1668 struct ib_wq **ind_tbl;
1671 enum port_pkey_state {
1672 IB_PORT_PKEY_NOT_VALID = 0,
1673 IB_PORT_PKEY_VALID = 1,
1674 IB_PORT_PKEY_LISTED = 2,
1677 struct ib_qp_security;
1679 struct ib_port_pkey {
1680 enum port_pkey_state state;
1683 struct list_head qp_list;
1684 struct list_head to_error_list;
1685 struct ib_qp_security *sec;
1688 struct ib_ports_pkeys {
1689 struct ib_port_pkey main;
1690 struct ib_port_pkey alt;
1693 struct ib_qp_security {
1695 struct ib_device *dev;
1696 /* Hold this mutex when changing port and pkey settings. */
1698 struct ib_ports_pkeys *ports_pkeys;
1699 /* A list of all open shared QP handles. Required to enforce security
1700 * properly for all users of a shared QP.
1702 struct list_head shared_qp_list;
1705 atomic_t error_list_count;
1706 struct completion error_complete;
1707 int error_comps_pending;
1711 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1712 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1715 struct ib_device *device;
1717 struct ib_cq *send_cq;
1718 struct ib_cq *recv_cq;
1721 struct list_head rdma_mrs;
1722 struct list_head sig_mrs;
1724 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1725 struct list_head xrcd_list;
1727 /* count times opened, mcast attaches, flow attaches */
1729 struct list_head open_list;
1730 struct ib_qp *real_qp;
1731 struct ib_uobject *uobject;
1732 void (*event_handler)(struct ib_event *, void *);
1734 /* sgid_attrs associated with the AV's */
1735 const struct ib_gid_attr *av_sgid_attr;
1736 const struct ib_gid_attr *alt_path_sgid_attr;
1740 enum ib_qp_type qp_type;
1741 struct ib_rwq_ind_table *rwq_ind_tbl;
1742 struct ib_qp_security *qp_sec;
1747 * Implementation details of the RDMA core, don't use in drivers:
1749 struct rdma_restrack_entry res;
1751 /* The counter the qp is bind to */
1752 struct rdma_counter *counter;
1756 struct ib_device *device;
1759 struct ib_uobject *uobject;
1764 struct ib_device *device;
1770 unsigned int page_size;
1771 enum ib_mr_type type;
1774 struct ib_uobject *uobject; /* user */
1775 struct list_head qp_entry; /* FR */
1779 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1781 * Implementation details of the RDMA core, don't use in drivers:
1783 struct rdma_restrack_entry res;
1787 struct ib_device *device;
1789 struct ib_uobject *uobject;
1791 enum ib_mw_type type;
1795 struct ib_device *device;
1797 struct list_head list;
1802 /* Supported steering options */
1803 enum ib_flow_attr_type {
1804 /* steering according to rule specifications */
1805 IB_FLOW_ATTR_NORMAL = 0x0,
1806 /* default unicast and multicast rule -
1807 * receive all Eth traffic which isn't steered to any QP
1809 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1810 /* default multicast rule -
1811 * receive all Eth multicast traffic which isn't steered to any QP
1813 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1814 /* sniffer rule - receive all port traffic */
1815 IB_FLOW_ATTR_SNIFFER = 0x3
1818 /* Supported steering header types */
1819 enum ib_flow_spec_type {
1821 IB_FLOW_SPEC_ETH = 0x20,
1822 IB_FLOW_SPEC_IB = 0x22,
1824 IB_FLOW_SPEC_IPV4 = 0x30,
1825 IB_FLOW_SPEC_IPV6 = 0x31,
1826 IB_FLOW_SPEC_ESP = 0x34,
1828 IB_FLOW_SPEC_TCP = 0x40,
1829 IB_FLOW_SPEC_UDP = 0x41,
1830 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1831 IB_FLOW_SPEC_GRE = 0x51,
1832 IB_FLOW_SPEC_MPLS = 0x60,
1833 IB_FLOW_SPEC_INNER = 0x100,
1835 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1836 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1837 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1838 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1840 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1841 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1843 /* Flow steering rule priority is set according to it's domain.
1844 * Lower domain value means higher priority.
1846 enum ib_flow_domain {
1847 IB_FLOW_DOMAIN_USER,
1848 IB_FLOW_DOMAIN_ETHTOOL,
1851 IB_FLOW_DOMAIN_NUM /* Must be last */
1854 enum ib_flow_flags {
1855 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1856 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1857 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1860 struct ib_flow_eth_filter {
1869 struct ib_flow_spec_eth {
1872 struct ib_flow_eth_filter val;
1873 struct ib_flow_eth_filter mask;
1876 struct ib_flow_ib_filter {
1883 struct ib_flow_spec_ib {
1886 struct ib_flow_ib_filter val;
1887 struct ib_flow_ib_filter mask;
1890 /* IPv4 header flags */
1891 enum ib_ipv4_flags {
1892 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1893 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1894 last have this flag set */
1897 struct ib_flow_ipv4_filter {
1908 struct ib_flow_spec_ipv4 {
1911 struct ib_flow_ipv4_filter val;
1912 struct ib_flow_ipv4_filter mask;
1915 struct ib_flow_ipv6_filter {
1926 struct ib_flow_spec_ipv6 {
1929 struct ib_flow_ipv6_filter val;
1930 struct ib_flow_ipv6_filter mask;
1933 struct ib_flow_tcp_udp_filter {
1940 struct ib_flow_spec_tcp_udp {
1943 struct ib_flow_tcp_udp_filter val;
1944 struct ib_flow_tcp_udp_filter mask;
1947 struct ib_flow_tunnel_filter {
1952 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1953 * the tunnel_id from val has the vni value
1955 struct ib_flow_spec_tunnel {
1958 struct ib_flow_tunnel_filter val;
1959 struct ib_flow_tunnel_filter mask;
1962 struct ib_flow_esp_filter {
1969 struct ib_flow_spec_esp {
1972 struct ib_flow_esp_filter val;
1973 struct ib_flow_esp_filter mask;
1976 struct ib_flow_gre_filter {
1977 __be16 c_ks_res0_ver;
1984 struct ib_flow_spec_gre {
1987 struct ib_flow_gre_filter val;
1988 struct ib_flow_gre_filter mask;
1991 struct ib_flow_mpls_filter {
1997 struct ib_flow_spec_mpls {
2000 struct ib_flow_mpls_filter val;
2001 struct ib_flow_mpls_filter mask;
2004 struct ib_flow_spec_action_tag {
2005 enum ib_flow_spec_type type;
2010 struct ib_flow_spec_action_drop {
2011 enum ib_flow_spec_type type;
2015 struct ib_flow_spec_action_handle {
2016 enum ib_flow_spec_type type;
2018 struct ib_flow_action *act;
2021 enum ib_counters_description {
2026 struct ib_flow_spec_action_count {
2027 enum ib_flow_spec_type type;
2029 struct ib_counters *counters;
2032 union ib_flow_spec {
2037 struct ib_flow_spec_eth eth;
2038 struct ib_flow_spec_ib ib;
2039 struct ib_flow_spec_ipv4 ipv4;
2040 struct ib_flow_spec_tcp_udp tcp_udp;
2041 struct ib_flow_spec_ipv6 ipv6;
2042 struct ib_flow_spec_tunnel tunnel;
2043 struct ib_flow_spec_esp esp;
2044 struct ib_flow_spec_gre gre;
2045 struct ib_flow_spec_mpls mpls;
2046 struct ib_flow_spec_action_tag flow_tag;
2047 struct ib_flow_spec_action_drop drop;
2048 struct ib_flow_spec_action_handle action;
2049 struct ib_flow_spec_action_count flow_count;
2052 struct ib_flow_attr {
2053 enum ib_flow_attr_type type;
2059 union ib_flow_spec flows[];
2064 struct ib_device *device;
2065 struct ib_uobject *uobject;
2068 enum ib_flow_action_type {
2069 IB_FLOW_ACTION_UNSPECIFIED,
2070 IB_FLOW_ACTION_ESP = 1,
2073 struct ib_flow_action_attrs_esp_keymats {
2074 enum ib_uverbs_flow_action_esp_keymat protocol;
2076 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2080 struct ib_flow_action_attrs_esp_replays {
2081 enum ib_uverbs_flow_action_esp_replay protocol;
2083 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2087 enum ib_flow_action_attrs_esp_flags {
2088 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2089 * This is done in order to share the same flags between user-space and
2090 * kernel and spare an unnecessary translation.
2094 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2095 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2098 struct ib_flow_spec_list {
2099 struct ib_flow_spec_list *next;
2100 union ib_flow_spec spec;
2103 struct ib_flow_action_attrs_esp {
2104 struct ib_flow_action_attrs_esp_keymats *keymat;
2105 struct ib_flow_action_attrs_esp_replays *replay;
2106 struct ib_flow_spec_list *encap;
2107 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2108 * Value of 0 is a valid value.
2114 /* Use enum ib_flow_action_attrs_esp_flags */
2116 u64 hard_limit_pkts;
2119 struct ib_flow_action {
2120 struct ib_device *device;
2121 struct ib_uobject *uobject;
2122 enum ib_flow_action_type type;
2129 enum ib_process_mad_flags {
2130 IB_MAD_IGNORE_MKEY = 1,
2131 IB_MAD_IGNORE_BKEY = 2,
2132 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2135 enum ib_mad_result {
2136 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2137 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2138 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2139 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2142 struct ib_port_cache {
2144 struct ib_pkey_cache *pkey;
2145 struct ib_gid_table *gid;
2147 enum ib_port_state port_state;
2152 struct ib_event_handler event_handler;
2155 struct ib_port_immutable {
2162 struct ib_port_data {
2163 struct ib_device *ib_dev;
2165 struct ib_port_immutable immutable;
2167 spinlock_t pkey_list_lock;
2168 struct list_head pkey_list;
2170 struct ib_port_cache cache;
2172 spinlock_t netdev_lock;
2173 struct net_device __rcu *netdev;
2174 struct hlist_node ndev_hash_link;
2175 struct rdma_port_counter port_counter;
2176 struct rdma_hw_stats *hw_stats;
2179 /* rdma netdev type - specifies protocol type */
2180 enum rdma_netdev_t {
2181 RDMA_NETDEV_OPA_VNIC,
2186 * struct rdma_netdev - rdma netdev
2187 * For cases where netstack interfacing is required.
2189 struct rdma_netdev {
2191 struct ib_device *hca;
2195 * cleanup function must be specified.
2196 * FIXME: This is only used for OPA_VNIC and that usage should be
2199 void (*free_rdma_netdev)(struct net_device *netdev);
2201 /* control functions */
2202 void (*set_id)(struct net_device *netdev, int id);
2204 int (*send)(struct net_device *dev, struct sk_buff *skb,
2205 struct ib_ah *address, u32 dqpn);
2207 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2208 union ib_gid *gid, u16 mlid,
2209 int set_qkey, u32 qkey);
2210 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2211 union ib_gid *gid, u16 mlid);
2214 struct rdma_netdev_alloc_params {
2220 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2221 struct net_device *netdev, void *param);
2224 struct ib_odp_counters {
2226 atomic64_t invalidations;
2229 struct ib_counters {
2230 struct ib_device *device;
2231 struct ib_uobject *uobject;
2232 /* num of objects attached */
2236 struct ib_counters_read_attr {
2239 u32 flags; /* use enum ib_read_counters_flags */
2242 struct uverbs_attr_bundle;
2244 struct iw_cm_conn_param;
2246 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2247 .size_##ib_struct = \
2248 (sizeof(struct drv_struct) + \
2249 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2250 BUILD_BUG_ON_ZERO( \
2251 !__same_type(((struct drv_struct *)NULL)->member, \
2254 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2255 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2257 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2258 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2260 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2262 struct rdma_user_mmap_entry {
2264 struct ib_ucontext *ucontext;
2265 unsigned long start_pgoff;
2267 bool driver_removed;
2270 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2272 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2274 return (u64)entry->start_pgoff << PAGE_SHIFT;
2278 * struct ib_device_ops - InfiniBand device operations
2279 * This structure defines all the InfiniBand device operations, providers will
2280 * need to define the supported operations, otherwise they will be set to null.
2282 struct ib_device_ops {
2283 struct module *owner;
2284 enum rdma_driver_id driver_id;
2286 unsigned int uverbs_no_driver_id_binding:1;
2288 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2289 const struct ib_send_wr **bad_send_wr);
2290 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2291 const struct ib_recv_wr **bad_recv_wr);
2292 void (*drain_rq)(struct ib_qp *qp);
2293 void (*drain_sq)(struct ib_qp *qp);
2294 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2295 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2296 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2297 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2298 int (*post_srq_recv)(struct ib_srq *srq,
2299 const struct ib_recv_wr *recv_wr,
2300 const struct ib_recv_wr **bad_recv_wr);
2301 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2302 u8 port_num, const struct ib_wc *in_wc,
2303 const struct ib_grh *in_grh,
2304 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2305 size_t *out_mad_size, u16 *out_mad_pkey_index);
2306 int (*query_device)(struct ib_device *device,
2307 struct ib_device_attr *device_attr,
2308 struct ib_udata *udata);
2309 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2310 struct ib_device_modify *device_modify);
2311 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2312 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2314 int (*query_port)(struct ib_device *device, u8 port_num,
2315 struct ib_port_attr *port_attr);
2316 int (*modify_port)(struct ib_device *device, u8 port_num,
2317 int port_modify_mask,
2318 struct ib_port_modify *port_modify);
2320 * The following mandatory functions are used only at device
2321 * registration. Keep functions such as these at the end of this
2322 * structure to avoid cache line misses when accessing struct ib_device
2325 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2326 struct ib_port_immutable *immutable);
2327 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2330 * When calling get_netdev, the HW vendor's driver should return the
2331 * net device of device @device at port @port_num or NULL if such
2332 * a net device doesn't exist. The vendor driver should call dev_hold
2333 * on this net device. The HW vendor's device driver must guarantee
2334 * that this function returns NULL before the net device has finished
2335 * NETDEV_UNREGISTER state.
2337 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2339 * rdma netdev operation
2341 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2342 * must return -EOPNOTSUPP if it doesn't support the specified type.
2344 struct net_device *(*alloc_rdma_netdev)(
2345 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2346 const char *name, unsigned char name_assign_type,
2347 void (*setup)(struct net_device *));
2349 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2350 enum rdma_netdev_t type,
2351 struct rdma_netdev_alloc_params *params);
2353 * query_gid should be return GID value for @device, when @port_num
2354 * link layer is either IB or iWarp. It is no-op if @port_num port
2355 * is RoCE link layer.
2357 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2360 * When calling add_gid, the HW vendor's driver should add the gid
2361 * of device of port at gid index available at @attr. Meta-info of
2362 * that gid (for example, the network device related to this gid) is
2363 * available at @attr. @context allows the HW vendor driver to store
2364 * extra information together with a GID entry. The HW vendor driver may
2365 * allocate memory to contain this information and store it in @context
2366 * when a new GID entry is written to. Params are consistent until the
2367 * next call of add_gid or delete_gid. The function should return 0 on
2368 * success or error otherwise. The function could be called
2369 * concurrently for different ports. This function is only called when
2370 * roce_gid_table is used.
2372 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2374 * When calling del_gid, the HW vendor's driver should delete the
2375 * gid of device @device at gid index gid_index of port port_num
2376 * available in @attr.
2377 * Upon the deletion of a GID entry, the HW vendor must free any
2378 * allocated memory. The caller will clear @context afterwards.
2379 * This function is only called when roce_gid_table is used.
2381 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2382 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2384 int (*alloc_ucontext)(struct ib_ucontext *context,
2385 struct ib_udata *udata);
2386 void (*dealloc_ucontext)(struct ib_ucontext *context);
2387 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2389 * This will be called once refcount of an entry in mmap_xa reaches
2390 * zero. The type of the memory that was mapped may differ between
2391 * entries and is opaque to the rdma_user_mmap interface.
2392 * Therefore needs to be implemented by the driver in mmap_free.
2394 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2395 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2396 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2397 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2398 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2399 u32 flags, struct ib_udata *udata);
2400 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2401 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2402 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2403 int (*create_srq)(struct ib_srq *srq,
2404 struct ib_srq_init_attr *srq_init_attr,
2405 struct ib_udata *udata);
2406 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2407 enum ib_srq_attr_mask srq_attr_mask,
2408 struct ib_udata *udata);
2409 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2410 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2411 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2412 struct ib_qp_init_attr *qp_init_attr,
2413 struct ib_udata *udata);
2414 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2415 int qp_attr_mask, struct ib_udata *udata);
2416 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2417 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2418 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2419 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2420 struct ib_udata *udata);
2421 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2422 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2423 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2424 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2425 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2426 u64 virt_addr, int mr_access_flags,
2427 struct ib_udata *udata);
2428 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2429 u64 virt_addr, int mr_access_flags,
2430 struct ib_pd *pd, struct ib_udata *udata);
2431 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2432 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2433 u32 max_num_sg, struct ib_udata *udata);
2434 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2435 u32 max_num_data_sg,
2436 u32 max_num_meta_sg);
2437 int (*advise_mr)(struct ib_pd *pd,
2438 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2439 struct ib_sge *sg_list, u32 num_sge,
2440 struct uverbs_attr_bundle *attrs);
2441 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2442 unsigned int *sg_offset);
2443 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2444 struct ib_mr_status *mr_status);
2445 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2446 struct ib_udata *udata);
2447 int (*dealloc_mw)(struct ib_mw *mw);
2448 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2449 struct ib_fmr_attr *fmr_attr);
2450 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2452 int (*unmap_fmr)(struct list_head *fmr_list);
2453 int (*dealloc_fmr)(struct ib_fmr *fmr);
2454 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2455 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2456 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2457 struct ib_udata *udata);
2458 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2459 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2460 struct ib_flow_attr *flow_attr,
2461 int domain, struct ib_udata *udata);
2462 int (*destroy_flow)(struct ib_flow *flow_id);
2463 struct ib_flow_action *(*create_flow_action_esp)(
2464 struct ib_device *device,
2465 const struct ib_flow_action_attrs_esp *attr,
2466 struct uverbs_attr_bundle *attrs);
2467 int (*destroy_flow_action)(struct ib_flow_action *action);
2468 int (*modify_flow_action_esp)(
2469 struct ib_flow_action *action,
2470 const struct ib_flow_action_attrs_esp *attr,
2471 struct uverbs_attr_bundle *attrs);
2472 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2474 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2475 struct ifla_vf_info *ivf);
2476 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2477 struct ifla_vf_stats *stats);
2478 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2479 struct ifla_vf_guid *node_guid,
2480 struct ifla_vf_guid *port_guid);
2481 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2483 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2484 struct ib_wq_init_attr *init_attr,
2485 struct ib_udata *udata);
2486 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2487 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2488 u32 wq_attr_mask, struct ib_udata *udata);
2489 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2490 struct ib_device *device,
2491 struct ib_rwq_ind_table_init_attr *init_attr,
2492 struct ib_udata *udata);
2493 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2494 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2495 struct ib_ucontext *context,
2496 struct ib_dm_alloc_attr *attr,
2497 struct uverbs_attr_bundle *attrs);
2498 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2499 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2500 struct ib_dm_mr_attr *attr,
2501 struct uverbs_attr_bundle *attrs);
2502 struct ib_counters *(*create_counters)(
2503 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2504 int (*destroy_counters)(struct ib_counters *counters);
2505 int (*read_counters)(struct ib_counters *counters,
2506 struct ib_counters_read_attr *counters_read_attr,
2507 struct uverbs_attr_bundle *attrs);
2508 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2509 int data_sg_nents, unsigned int *data_sg_offset,
2510 struct scatterlist *meta_sg, int meta_sg_nents,
2511 unsigned int *meta_sg_offset);
2514 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2515 * driver initialized data. The struct is kfree()'ed by the sysfs
2516 * core when the device is removed. A lifespan of -1 in the return
2517 * struct tells the core to set a default lifespan.
2519 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2522 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2523 * @index - The index in the value array we wish to have updated, or
2524 * num_counters if we want all stats updated
2526 * < 0 - Error, no counters updated
2527 * index - Updated the single counter pointed to by index
2528 * num_counters - Updated all counters (will reset the timestamp
2529 * and prevent further calls for lifespan milliseconds)
2530 * Drivers are allowed to update all counters in leiu of just the
2531 * one given in index at their option
2533 int (*get_hw_stats)(struct ib_device *device,
2534 struct rdma_hw_stats *stats, u8 port, int index);
2536 * This function is called once for each port when a ib device is
2539 int (*init_port)(struct ib_device *device, u8 port_num,
2540 struct kobject *port_sysfs);
2542 * Allows rdma drivers to add their own restrack attributes.
2544 int (*fill_res_entry)(struct sk_buff *msg,
2545 struct rdma_restrack_entry *entry);
2547 /* Device lifecycle callbacks */
2549 * Called after the device becomes registered, before clients are
2552 int (*enable_driver)(struct ib_device *dev);
2554 * This is called as part of ib_dealloc_device().
2556 void (*dealloc_driver)(struct ib_device *dev);
2558 /* iWarp CM callbacks */
2559 void (*iw_add_ref)(struct ib_qp *qp);
2560 void (*iw_rem_ref)(struct ib_qp *qp);
2561 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2562 int (*iw_connect)(struct iw_cm_id *cm_id,
2563 struct iw_cm_conn_param *conn_param);
2564 int (*iw_accept)(struct iw_cm_id *cm_id,
2565 struct iw_cm_conn_param *conn_param);
2566 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2568 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2569 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2571 * counter_bind_qp - Bind a QP to a counter.
2572 * @counter - The counter to be bound. If counter->id is zero then
2573 * the driver needs to allocate a new counter and set counter->id
2575 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2577 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2578 * counter and bind it onto the default one
2580 int (*counter_unbind_qp)(struct ib_qp *qp);
2582 * counter_dealloc -De-allocate the hw counter
2584 int (*counter_dealloc)(struct rdma_counter *counter);
2586 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2587 * the driver initialized data.
2589 struct rdma_hw_stats *(*counter_alloc_stats)(
2590 struct rdma_counter *counter);
2592 * counter_update_stats - Query the stats value of this counter
2594 int (*counter_update_stats)(struct rdma_counter *counter);
2597 * Allows rdma drivers to add their own restrack attributes
2598 * dumped via 'rdma stat' iproute2 command.
2600 int (*fill_stat_entry)(struct sk_buff *msg,
2601 struct rdma_restrack_entry *entry);
2603 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2604 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2605 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2606 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2607 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2610 struct ib_core_device {
2611 /* device must be the first element in structure until,
2612 * union of ib_core_device and device exists in ib_device.
2615 possible_net_t rdma_net;
2616 struct kobject *ports_kobj;
2617 struct list_head port_list;
2618 struct ib_device *owner; /* reach back to owner ib_device */
2621 struct rdma_restrack_root;
2623 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2624 struct device *dma_device;
2625 struct ib_device_ops ops;
2626 char name[IB_DEVICE_NAME_MAX];
2627 struct rcu_head rcu_head;
2629 struct list_head event_handler_list;
2630 spinlock_t event_handler_lock;
2632 struct rw_semaphore client_data_rwsem;
2633 struct xarray client_data;
2634 struct mutex unregistration_lock;
2636 struct ib_cache cache;
2638 * port_data is indexed by port number
2640 struct ib_port_data *port_data;
2642 int num_comp_vectors;
2646 struct ib_core_device coredev;
2649 /* First group for device attributes,
2650 * Second group for driver provided attributes (optional).
2651 * It is NULL terminated array.
2653 const struct attribute_group *groups[3];
2655 u64 uverbs_cmd_mask;
2656 u64 uverbs_ex_cmd_mask;
2658 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2662 /* Indicates kernel verbs support, should not be used in drivers */
2663 u16 kverbs_provider:1;
2664 /* CQ adaptive moderation (RDMA DIM) */
2668 struct ib_device_attr attrs;
2669 struct attribute_group *hw_stats_ag;
2670 struct rdma_hw_stats *hw_stats;
2672 #ifdef CONFIG_CGROUP_RDMA
2673 struct rdmacg_device cg_device;
2677 struct rdma_restrack_root *res;
2679 const struct uapi_definition *driver_def;
2682 * Positive refcount indicates that the device is currently
2683 * registered and cannot be unregistered.
2685 refcount_t refcount;
2686 struct completion unreg_completion;
2687 struct work_struct unregistration_work;
2689 const struct rdma_link_ops *link_ops;
2691 /* Protects compat_devs xarray modifications */
2692 struct mutex compat_devs_mutex;
2693 /* Maintains compat devices for each net namespace */
2694 struct xarray compat_devs;
2696 /* Used by iWarp CM */
2697 char iw_ifname[IFNAMSIZ];
2698 u32 iw_driver_flags;
2701 struct ib_client_nl_info;
2704 void (*add) (struct ib_device *);
2705 void (*remove)(struct ib_device *, void *client_data);
2706 void (*rename)(struct ib_device *dev, void *client_data);
2707 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2708 struct ib_client_nl_info *res);
2709 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2711 /* Returns the net_dev belonging to this ib_client and matching the
2713 * @dev: An RDMA device that the net_dev use for communication.
2714 * @port: A physical port number on the RDMA device.
2715 * @pkey: P_Key that the net_dev uses if applicable.
2716 * @gid: A GID that the net_dev uses to communicate.
2717 * @addr: An IP address the net_dev is configured with.
2718 * @client_data: The device's client data set by ib_set_client_data().
2720 * An ib_client that implements a net_dev on top of RDMA devices
2721 * (such as IP over IB) should implement this callback, allowing the
2722 * rdma_cm module to find the right net_dev for a given request.
2724 * The caller is responsible for calling dev_put on the returned
2726 struct net_device *(*get_net_dev_by_params)(
2727 struct ib_device *dev,
2730 const union ib_gid *gid,
2731 const struct sockaddr *addr,
2735 struct completion uses_zero;
2738 /* kverbs are not required by the client */
2743 * IB block DMA iterator
2745 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2746 * to a HW supported page size.
2748 struct ib_block_iter {
2749 /* internal states */
2750 struct scatterlist *__sg; /* sg holding the current aligned block */
2751 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2752 unsigned int __sg_nents; /* number of SG entries */
2753 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2754 unsigned int __pg_bit; /* alignment of current block */
2757 struct ib_device *_ib_alloc_device(size_t size);
2758 #define ib_alloc_device(drv_struct, member) \
2759 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2760 BUILD_BUG_ON_ZERO(offsetof( \
2761 struct drv_struct, member))), \
2762 struct drv_struct, member)
2764 void ib_dealloc_device(struct ib_device *device);
2766 void ib_get_device_fw_str(struct ib_device *device, char *str);
2768 int ib_register_device(struct ib_device *device, const char *name);
2769 void ib_unregister_device(struct ib_device *device);
2770 void ib_unregister_driver(enum rdma_driver_id driver_id);
2771 void ib_unregister_device_and_put(struct ib_device *device);
2772 void ib_unregister_device_queued(struct ib_device *ib_dev);
2774 int ib_register_client (struct ib_client *client);
2775 void ib_unregister_client(struct ib_client *client);
2777 void __rdma_block_iter_start(struct ib_block_iter *biter,
2778 struct scatterlist *sglist,
2780 unsigned long pgsz);
2781 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2784 * rdma_block_iter_dma_address - get the aligned dma address of the current
2785 * block held by the block iterator.
2786 * @biter: block iterator holding the memory block
2788 static inline dma_addr_t
2789 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2791 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2795 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2796 * @sglist: sglist to iterate over
2797 * @biter: block iterator holding the memory block
2798 * @nents: maximum number of sg entries to iterate over
2799 * @pgsz: best HW supported page size to use
2801 * Callers may use rdma_block_iter_dma_address() to get each
2802 * blocks aligned DMA address.
2804 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2805 for (__rdma_block_iter_start(biter, sglist, nents, \
2807 __rdma_block_iter_next(biter);)
2810 * ib_get_client_data - Get IB client context
2811 * @device:Device to get context for
2812 * @client:Client to get context for
2814 * ib_get_client_data() returns the client context data set with
2815 * ib_set_client_data(). This can only be called while the client is
2816 * registered to the device, once the ib_client remove() callback returns this
2819 static inline void *ib_get_client_data(struct ib_device *device,
2820 struct ib_client *client)
2822 return xa_load(&device->client_data, client->client_id);
2824 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2826 void ib_set_device_ops(struct ib_device *device,
2827 const struct ib_device_ops *ops);
2829 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2830 unsigned long pfn, unsigned long size, pgprot_t prot,
2831 struct rdma_user_mmap_entry *entry);
2832 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2833 struct rdma_user_mmap_entry *entry,
2835 struct rdma_user_mmap_entry *
2836 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2837 unsigned long pgoff);
2838 struct rdma_user_mmap_entry *
2839 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2840 struct vm_area_struct *vma);
2841 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2843 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2845 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2847 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2850 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2852 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2855 static inline bool ib_is_buffer_cleared(const void __user *p,
2861 if (len > USHRT_MAX)
2864 buf = memdup_user(p, len);
2868 ret = !memchr_inv(buf, 0, len);
2873 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2877 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2881 * ib_is_destroy_retryable - Check whether the uobject destruction
2883 * @ret: The initial destruction return code
2884 * @why: remove reason
2885 * @uobj: The uobject that is destroyed
2887 * This function is a helper function that IB layer and low-level drivers
2888 * can use to consider whether the destruction of the given uobject is
2890 * It checks the original return code, if it wasn't success the destruction
2891 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2892 * the remove reason. (i.e. why).
2893 * Must be called with the object locked for destroy.
2895 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2896 struct ib_uobject *uobj)
2898 return ret && (why == RDMA_REMOVE_DESTROY ||
2899 uobj->context->cleanup_retryable);
2903 * ib_destroy_usecnt - Called during destruction to check the usecnt
2904 * @usecnt: The usecnt atomic
2905 * @why: remove reason
2906 * @uobj: The uobject that is destroyed
2908 * Non-zero usecnts will block destruction unless destruction was triggered by
2909 * a ucontext cleanup.
2911 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2912 enum rdma_remove_reason why,
2913 struct ib_uobject *uobj)
2915 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2921 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2922 * contains all required attributes and no attributes not allowed for
2923 * the given QP state transition.
2924 * @cur_state: Current QP state
2925 * @next_state: Next QP state
2927 * @mask: Mask of supplied QP attributes
2929 * This function is a helper function that a low-level driver's
2930 * modify_qp method can use to validate the consumer's input. It
2931 * checks that cur_state and next_state are valid QP states, that a
2932 * transition from cur_state to next_state is allowed by the IB spec,
2933 * and that the attribute mask supplied is allowed for the transition.
2935 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2936 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2938 void ib_register_event_handler(struct ib_event_handler *event_handler);
2939 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2940 void ib_dispatch_event(struct ib_event *event);
2942 int ib_query_port(struct ib_device *device,
2943 u8 port_num, struct ib_port_attr *port_attr);
2945 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2949 * rdma_cap_ib_switch - Check if the device is IB switch
2950 * @device: Device to check
2952 * Device driver is responsible for setting is_switch bit on
2953 * in ib_device structure at init time.
2955 * Return: true if the device is IB switch.
2957 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2959 return device->is_switch;
2963 * rdma_start_port - Return the first valid port number for the device
2966 * @device: Device to be checked
2968 * Return start port number
2970 static inline u8 rdma_start_port(const struct ib_device *device)
2972 return rdma_cap_ib_switch(device) ? 0 : 1;
2976 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2977 * @device - The struct ib_device * to iterate over
2978 * @iter - The unsigned int to store the port number
2980 #define rdma_for_each_port(device, iter) \
2981 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2982 unsigned int, iter))); \
2983 iter <= rdma_end_port(device); (iter)++)
2986 * rdma_end_port - Return the last valid port number for the device
2989 * @device: Device to be checked
2991 * Return last port number
2993 static inline u8 rdma_end_port(const struct ib_device *device)
2995 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2998 static inline int rdma_is_port_valid(const struct ib_device *device,
3001 return (port >= rdma_start_port(device) &&
3002 port <= rdma_end_port(device));
3005 static inline bool rdma_is_grh_required(const struct ib_device *device,
3008 return device->port_data[port_num].immutable.core_cap_flags &
3009 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3012 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3014 return device->port_data[port_num].immutable.core_cap_flags &
3015 RDMA_CORE_CAP_PROT_IB;
3018 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3020 return device->port_data[port_num].immutable.core_cap_flags &
3021 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3024 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3026 return device->port_data[port_num].immutable.core_cap_flags &
3027 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3030 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3032 return device->port_data[port_num].immutable.core_cap_flags &
3033 RDMA_CORE_CAP_PROT_ROCE;
3036 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3038 return device->port_data[port_num].immutable.core_cap_flags &
3039 RDMA_CORE_CAP_PROT_IWARP;
3042 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3044 return rdma_protocol_ib(device, port_num) ||
3045 rdma_protocol_roce(device, port_num);
3048 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3050 return device->port_data[port_num].immutable.core_cap_flags &
3051 RDMA_CORE_CAP_PROT_RAW_PACKET;
3054 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3056 return device->port_data[port_num].immutable.core_cap_flags &
3057 RDMA_CORE_CAP_PROT_USNIC;
3061 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3062 * Management Datagrams.
3063 * @device: Device to check
3064 * @port_num: Port number to check
3066 * Management Datagrams (MAD) are a required part of the InfiniBand
3067 * specification and are supported on all InfiniBand devices. A slightly
3068 * extended version are also supported on OPA interfaces.
3070 * Return: true if the port supports sending/receiving of MAD packets.
3072 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3074 return device->port_data[port_num].immutable.core_cap_flags &
3075 RDMA_CORE_CAP_IB_MAD;
3079 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3080 * Management Datagrams.
3081 * @device: Device to check
3082 * @port_num: Port number to check
3084 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3085 * datagrams with their own versions. These OPA MADs share many but not all of
3086 * the characteristics of InfiniBand MADs.
3088 * OPA MADs differ in the following ways:
3090 * 1) MADs are variable size up to 2K
3091 * IBTA defined MADs remain fixed at 256 bytes
3092 * 2) OPA SMPs must carry valid PKeys
3093 * 3) OPA SMP packets are a different format
3095 * Return: true if the port supports OPA MAD packet formats.
3097 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3099 return device->port_data[port_num].immutable.core_cap_flags &
3100 RDMA_CORE_CAP_OPA_MAD;
3104 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3105 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3106 * @device: Device to check
3107 * @port_num: Port number to check
3109 * Each InfiniBand node is required to provide a Subnet Management Agent
3110 * that the subnet manager can access. Prior to the fabric being fully
3111 * configured by the subnet manager, the SMA is accessed via a well known
3112 * interface called the Subnet Management Interface (SMI). This interface
3113 * uses directed route packets to communicate with the SM to get around the
3114 * chicken and egg problem of the SM needing to know what's on the fabric
3115 * in order to configure the fabric, and needing to configure the fabric in
3116 * order to send packets to the devices on the fabric. These directed
3117 * route packets do not need the fabric fully configured in order to reach
3118 * their destination. The SMI is the only method allowed to send
3119 * directed route packets on an InfiniBand fabric.
3121 * Return: true if the port provides an SMI.
3123 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3125 return device->port_data[port_num].immutable.core_cap_flags &
3126 RDMA_CORE_CAP_IB_SMI;
3130 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3131 * Communication Manager.
3132 * @device: Device to check
3133 * @port_num: Port number to check
3135 * The InfiniBand Communication Manager is one of many pre-defined General
3136 * Service Agents (GSA) that are accessed via the General Service
3137 * Interface (GSI). It's role is to facilitate establishment of connections
3138 * between nodes as well as other management related tasks for established
3141 * Return: true if the port supports an IB CM (this does not guarantee that
3142 * a CM is actually running however).
3144 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3146 return device->port_data[port_num].immutable.core_cap_flags &
3147 RDMA_CORE_CAP_IB_CM;
3151 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3152 * Communication Manager.
3153 * @device: Device to check
3154 * @port_num: Port number to check
3156 * Similar to above, but specific to iWARP connections which have a different
3157 * managment protocol than InfiniBand.
3159 * Return: true if the port supports an iWARP CM (this does not guarantee that
3160 * a CM is actually running however).
3162 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3164 return device->port_data[port_num].immutable.core_cap_flags &
3165 RDMA_CORE_CAP_IW_CM;
3169 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3170 * Subnet Administration.
3171 * @device: Device to check
3172 * @port_num: Port number to check
3174 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3175 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3176 * fabrics, devices should resolve routes to other hosts by contacting the
3177 * SA to query the proper route.
3179 * Return: true if the port should act as a client to the fabric Subnet
3180 * Administration interface. This does not imply that the SA service is
3183 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3185 return device->port_data[port_num].immutable.core_cap_flags &
3186 RDMA_CORE_CAP_IB_SA;
3190 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3192 * @device: Device to check
3193 * @port_num: Port number to check
3195 * InfiniBand multicast registration is more complex than normal IPv4 or
3196 * IPv6 multicast registration. Each Host Channel Adapter must register
3197 * with the Subnet Manager when it wishes to join a multicast group. It
3198 * should do so only once regardless of how many queue pairs it subscribes
3199 * to this group. And it should leave the group only after all queue pairs
3200 * attached to the group have been detached.
3202 * Return: true if the port must undertake the additional adminstrative
3203 * overhead of registering/unregistering with the SM and tracking of the
3204 * total number of queue pairs attached to the multicast group.
3206 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3208 return rdma_cap_ib_sa(device, port_num);
3212 * rdma_cap_af_ib - Check if the port of device has the capability
3213 * Native Infiniband Address.
3214 * @device: Device to check
3215 * @port_num: Port number to check
3217 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3218 * GID. RoCE uses a different mechanism, but still generates a GID via
3219 * a prescribed mechanism and port specific data.
3221 * Return: true if the port uses a GID address to identify devices on the
3224 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3226 return device->port_data[port_num].immutable.core_cap_flags &
3227 RDMA_CORE_CAP_AF_IB;
3231 * rdma_cap_eth_ah - Check if the port of device has the capability
3232 * Ethernet Address Handle.
3233 * @device: Device to check
3234 * @port_num: Port number to check
3236 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3237 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3238 * port. Normally, packet headers are generated by the sending host
3239 * adapter, but when sending connectionless datagrams, we must manually
3240 * inject the proper headers for the fabric we are communicating over.
3242 * Return: true if we are running as a RoCE port and must force the
3243 * addition of a Global Route Header built from our Ethernet Address
3244 * Handle into our header list for connectionless packets.
3246 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3248 return device->port_data[port_num].immutable.core_cap_flags &
3249 RDMA_CORE_CAP_ETH_AH;
3253 * rdma_cap_opa_ah - Check if the port of device supports
3254 * OPA Address handles
3255 * @device: Device to check
3256 * @port_num: Port number to check
3258 * Return: true if we are running on an OPA device which supports
3259 * the extended OPA addressing.
3261 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3263 return (device->port_data[port_num].immutable.core_cap_flags &
3264 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3268 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3271 * @port_num: Port number
3273 * This MAD size includes the MAD headers and MAD payload. No other headers
3276 * Return the max MAD size required by the Port. Will return 0 if the port
3277 * does not support MADs
3279 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3281 return device->port_data[port_num].immutable.max_mad_size;
3285 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3286 * @device: Device to check
3287 * @port_num: Port number to check
3289 * RoCE GID table mechanism manages the various GIDs for a device.
3291 * NOTE: if allocating the port's GID table has failed, this call will still
3292 * return true, but any RoCE GID table API will fail.
3294 * Return: true if the port uses RoCE GID table mechanism in order to manage
3297 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3300 return rdma_protocol_roce(device, port_num) &&
3301 device->ops.add_gid && device->ops.del_gid;
3305 * Check if the device supports READ W/ INVALIDATE.
3307 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3310 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3311 * has support for it yet.
3313 return rdma_protocol_iwarp(dev, port_num);
3317 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3320 * @pgsz_bitmap: bitmap of HW supported page sizes
3322 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3323 unsigned long pgsz_bitmap)
3325 unsigned long align;
3328 align = addr & -addr;
3330 /* Find page bit such that addr is aligned to the highest supported
3333 pgsz = pgsz_bitmap & ~(-align << 1);
3335 return __ffs(pgsz_bitmap);
3340 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3342 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3343 struct ifla_vf_info *info);
3344 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3345 struct ifla_vf_stats *stats);
3346 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3347 struct ifla_vf_guid *node_guid,
3348 struct ifla_vf_guid *port_guid);
3349 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3352 int ib_query_pkey(struct ib_device *device,
3353 u8 port_num, u16 index, u16 *pkey);
3355 int ib_modify_device(struct ib_device *device,
3356 int device_modify_mask,
3357 struct ib_device_modify *device_modify);
3359 int ib_modify_port(struct ib_device *device,
3360 u8 port_num, int port_modify_mask,
3361 struct ib_port_modify *port_modify);
3363 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3364 u8 *port_num, u16 *index);
3366 int ib_find_pkey(struct ib_device *device,
3367 u8 port_num, u16 pkey, u16 *index);
3371 * Create a memory registration for all memory in the system and place
3372 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3373 * ULPs to avoid the overhead of dynamic MRs.
3375 * This flag is generally considered unsafe and must only be used in
3376 * extremly trusted environments. Every use of it will log a warning
3377 * in the kernel log.
3379 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3382 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3383 const char *caller);
3385 #define ib_alloc_pd(device, flags) \
3386 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3389 * ib_dealloc_pd_user - Deallocate kernel/user PD
3390 * @pd: The protection domain
3391 * @udata: Valid user data or NULL for kernel objects
3393 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3396 * ib_dealloc_pd - Deallocate kernel PD
3397 * @pd: The protection domain
3399 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3401 static inline void ib_dealloc_pd(struct ib_pd *pd)
3403 ib_dealloc_pd_user(pd, NULL);
3406 enum rdma_create_ah_flags {
3407 /* In a sleepable context */
3408 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3412 * rdma_create_ah - Creates an address handle for the given address vector.
3413 * @pd: The protection domain associated with the address handle.
3414 * @ah_attr: The attributes of the address vector.
3415 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3417 * The address handle is used to reference a local or global destination
3418 * in all UD QP post sends.
3420 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3424 * rdma_create_user_ah - Creates an address handle for the given address vector.
3425 * It resolves destination mac address for ah attribute of RoCE type.
3426 * @pd: The protection domain associated with the address handle.
3427 * @ah_attr: The attributes of the address vector.
3428 * @udata: pointer to user's input output buffer information need by
3431 * It returns 0 on success and returns appropriate error code on error.
3432 * The address handle is used to reference a local or global destination
3433 * in all UD QP post sends.
3435 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3436 struct rdma_ah_attr *ah_attr,
3437 struct ib_udata *udata);
3439 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3441 * @hdr: the L3 header to parse
3442 * @net_type: type of header to parse
3443 * @sgid: place to store source gid
3444 * @dgid: place to store destination gid
3446 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3447 enum rdma_network_type net_type,
3448 union ib_gid *sgid, union ib_gid *dgid);
3451 * ib_get_rdma_header_version - Get the header version
3452 * @hdr: the L3 header to parse
3454 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3457 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3459 * @device: Device on which the received message arrived.
3460 * @port_num: Port on which the received message arrived.
3461 * @wc: Work completion associated with the received message.
3462 * @grh: References the received global route header. This parameter is
3463 * ignored unless the work completion indicates that the GRH is valid.
3464 * @ah_attr: Returned attributes that can be used when creating an address
3465 * handle for replying to the message.
3466 * When ib_init_ah_attr_from_wc() returns success,
3467 * (a) for IB link layer it optionally contains a reference to SGID attribute
3468 * when GRH is present for IB link layer.
3469 * (b) for RoCE link layer it contains a reference to SGID attribute.
3470 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3471 * attributes which are initialized using ib_init_ah_attr_from_wc().
3474 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3475 const struct ib_wc *wc, const struct ib_grh *grh,
3476 struct rdma_ah_attr *ah_attr);
3479 * ib_create_ah_from_wc - Creates an address handle associated with the
3480 * sender of the specified work completion.
3481 * @pd: The protection domain associated with the address handle.
3482 * @wc: Work completion information associated with a received message.
3483 * @grh: References the received global route header. This parameter is
3484 * ignored unless the work completion indicates that the GRH is valid.
3485 * @port_num: The outbound port number to associate with the address.
3487 * The address handle is used to reference a local or global destination
3488 * in all UD QP post sends.
3490 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3491 const struct ib_grh *grh, u8 port_num);
3494 * rdma_modify_ah - Modifies the address vector associated with an address
3496 * @ah: The address handle to modify.
3497 * @ah_attr: The new address vector attributes to associate with the
3500 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3503 * rdma_query_ah - Queries the address vector associated with an address
3505 * @ah: The address handle to query.
3506 * @ah_attr: The address vector attributes associated with the address
3509 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3511 enum rdma_destroy_ah_flags {
3512 /* In a sleepable context */
3513 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3517 * rdma_destroy_ah_user - Destroys an address handle.
3518 * @ah: The address handle to destroy.
3519 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3520 * @udata: Valid user data or NULL for kernel objects
3522 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3525 * rdma_destroy_ah - Destroys an kernel address handle.
3526 * @ah: The address handle to destroy.
3527 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3529 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3531 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3533 return rdma_destroy_ah_user(ah, flags, NULL);
3537 * ib_create_srq - Creates a SRQ associated with the specified protection
3539 * @pd: The protection domain associated with the SRQ.
3540 * @srq_init_attr: A list of initial attributes required to create the
3541 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3542 * the actual capabilities of the created SRQ.
3544 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3545 * requested size of the SRQ, and set to the actual values allocated
3546 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3547 * will always be at least as large as the requested values.
3549 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3550 struct ib_srq_init_attr *srq_init_attr);
3553 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3554 * @srq: The SRQ to modify.
3555 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3556 * the current values of selected SRQ attributes are returned.
3557 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3558 * are being modified.
3560 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3561 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3562 * the number of receives queued drops below the limit.
3564 int ib_modify_srq(struct ib_srq *srq,
3565 struct ib_srq_attr *srq_attr,
3566 enum ib_srq_attr_mask srq_attr_mask);
3569 * ib_query_srq - Returns the attribute list and current values for the
3571 * @srq: The SRQ to query.
3572 * @srq_attr: The attributes of the specified SRQ.
3574 int ib_query_srq(struct ib_srq *srq,
3575 struct ib_srq_attr *srq_attr);
3578 * ib_destroy_srq_user - Destroys the specified SRQ.
3579 * @srq: The SRQ to destroy.
3580 * @udata: Valid user data or NULL for kernel objects
3582 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3585 * ib_destroy_srq - Destroys the specified kernel SRQ.
3586 * @srq: The SRQ to destroy.
3588 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3590 static inline int ib_destroy_srq(struct ib_srq *srq)
3592 return ib_destroy_srq_user(srq, NULL);
3596 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3597 * @srq: The SRQ to post the work request on.
3598 * @recv_wr: A list of work requests to post on the receive queue.
3599 * @bad_recv_wr: On an immediate failure, this parameter will reference
3600 * the work request that failed to be posted on the QP.
3602 static inline int ib_post_srq_recv(struct ib_srq *srq,
3603 const struct ib_recv_wr *recv_wr,
3604 const struct ib_recv_wr **bad_recv_wr)
3606 const struct ib_recv_wr *dummy;
3608 return srq->device->ops.post_srq_recv(srq, recv_wr,
3609 bad_recv_wr ? : &dummy);
3613 * ib_create_qp_user - Creates a QP associated with the specified protection
3615 * @pd: The protection domain associated with the QP.
3616 * @qp_init_attr: A list of initial attributes required to create the
3617 * QP. If QP creation succeeds, then the attributes are updated to
3618 * the actual capabilities of the created QP.
3619 * @udata: Valid user data or NULL for kernel objects
3621 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3622 struct ib_qp_init_attr *qp_init_attr,
3623 struct ib_udata *udata);
3626 * ib_create_qp - Creates a kernel QP associated with the specified protection
3628 * @pd: The protection domain associated with the QP.
3629 * @qp_init_attr: A list of initial attributes required to create the
3630 * QP. If QP creation succeeds, then the attributes are updated to
3631 * the actual capabilities of the created QP.
3632 * @udata: Valid user data or NULL for kernel objects
3634 * NOTE: for user qp use ib_create_qp_user with valid udata!
3636 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3637 struct ib_qp_init_attr *qp_init_attr)
3639 return ib_create_qp_user(pd, qp_init_attr, NULL);
3643 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3644 * @qp: The QP to modify.
3645 * @attr: On input, specifies the QP attributes to modify. On output,
3646 * the current values of selected QP attributes are returned.
3647 * @attr_mask: A bit-mask used to specify which attributes of the QP
3648 * are being modified.
3649 * @udata: pointer to user's input output buffer information
3650 * are being modified.
3651 * It returns 0 on success and returns appropriate error code on error.
3653 int ib_modify_qp_with_udata(struct ib_qp *qp,
3654 struct ib_qp_attr *attr,
3656 struct ib_udata *udata);
3659 * ib_modify_qp - Modifies the attributes for the specified QP and then
3660 * transitions the QP to the given state.
3661 * @qp: The QP to modify.
3662 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3663 * the current values of selected QP attributes are returned.
3664 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3665 * are being modified.
3667 int ib_modify_qp(struct ib_qp *qp,
3668 struct ib_qp_attr *qp_attr,
3672 * ib_query_qp - Returns the attribute list and current values for the
3674 * @qp: The QP to query.
3675 * @qp_attr: The attributes of the specified QP.
3676 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3677 * @qp_init_attr: Additional attributes of the selected QP.
3679 * The qp_attr_mask may be used to limit the query to gathering only the
3680 * selected attributes.
3682 int ib_query_qp(struct ib_qp *qp,
3683 struct ib_qp_attr *qp_attr,
3685 struct ib_qp_init_attr *qp_init_attr);
3688 * ib_destroy_qp - Destroys the specified QP.
3689 * @qp: The QP to destroy.
3690 * @udata: Valid udata or NULL for kernel objects
3692 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3695 * ib_destroy_qp - Destroys the specified kernel QP.
3696 * @qp: The QP to destroy.
3698 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3700 static inline int ib_destroy_qp(struct ib_qp *qp)
3702 return ib_destroy_qp_user(qp, NULL);
3706 * ib_open_qp - Obtain a reference to an existing sharable QP.
3707 * @xrcd - XRC domain
3708 * @qp_open_attr: Attributes identifying the QP to open.
3710 * Returns a reference to a sharable QP.
3712 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3713 struct ib_qp_open_attr *qp_open_attr);
3716 * ib_close_qp - Release an external reference to a QP.
3717 * @qp: The QP handle to release
3719 * The opened QP handle is released by the caller. The underlying
3720 * shared QP is not destroyed until all internal references are released.
3722 int ib_close_qp(struct ib_qp *qp);
3725 * ib_post_send - Posts a list of work requests to the send queue of
3727 * @qp: The QP to post the work request on.
3728 * @send_wr: A list of work requests to post on the send queue.
3729 * @bad_send_wr: On an immediate failure, this parameter will reference
3730 * the work request that failed to be posted on the QP.
3732 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3733 * error is returned, the QP state shall not be affected,
3734 * ib_post_send() will return an immediate error after queueing any
3735 * earlier work requests in the list.
3737 static inline int ib_post_send(struct ib_qp *qp,
3738 const struct ib_send_wr *send_wr,
3739 const struct ib_send_wr **bad_send_wr)
3741 const struct ib_send_wr *dummy;
3743 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3747 * ib_post_recv - Posts a list of work requests to the receive queue of
3749 * @qp: The QP to post the work request on.
3750 * @recv_wr: A list of work requests to post on the receive queue.
3751 * @bad_recv_wr: On an immediate failure, this parameter will reference
3752 * the work request that failed to be posted on the QP.
3754 static inline int ib_post_recv(struct ib_qp *qp,
3755 const struct ib_recv_wr *recv_wr,
3756 const struct ib_recv_wr **bad_recv_wr)
3758 const struct ib_recv_wr *dummy;
3760 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3763 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3764 int nr_cqe, int comp_vector,
3765 enum ib_poll_context poll_ctx,
3766 const char *caller, struct ib_udata *udata);
3769 * ib_alloc_cq_user: Allocate kernel/user CQ
3770 * @dev: The IB device
3771 * @private: Private data attached to the CQE
3772 * @nr_cqe: Number of CQEs in the CQ
3773 * @comp_vector: Completion vector used for the IRQs
3774 * @poll_ctx: Context used for polling the CQ
3775 * @udata: Valid user data or NULL for kernel objects
3777 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3778 void *private, int nr_cqe,
3780 enum ib_poll_context poll_ctx,
3781 struct ib_udata *udata)
3783 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3784 KBUILD_MODNAME, udata);
3788 * ib_alloc_cq: Allocate kernel CQ
3789 * @dev: The IB device
3790 * @private: Private data attached to the CQE
3791 * @nr_cqe: Number of CQEs in the CQ
3792 * @comp_vector: Completion vector used for the IRQs
3793 * @poll_ctx: Context used for polling the CQ
3795 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3797 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3798 int nr_cqe, int comp_vector,
3799 enum ib_poll_context poll_ctx)
3801 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3805 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3806 int nr_cqe, enum ib_poll_context poll_ctx,
3807 const char *caller);
3810 * ib_alloc_cq_any: Allocate kernel CQ
3811 * @dev: The IB device
3812 * @private: Private data attached to the CQE
3813 * @nr_cqe: Number of CQEs in the CQ
3814 * @poll_ctx: Context used for polling the CQ
3816 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3817 void *private, int nr_cqe,
3818 enum ib_poll_context poll_ctx)
3820 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3825 * ib_free_cq_user - Free kernel/user CQ
3826 * @cq: The CQ to free
3827 * @udata: Valid user data or NULL for kernel objects
3829 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3832 * ib_free_cq - Free kernel CQ
3833 * @cq: The CQ to free
3835 * NOTE: for user cq use ib_free_cq_user with valid udata!
3837 static inline void ib_free_cq(struct ib_cq *cq)
3839 ib_free_cq_user(cq, NULL);
3842 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3845 * ib_create_cq - Creates a CQ on the specified device.
3846 * @device: The device on which to create the CQ.
3847 * @comp_handler: A user-specified callback that is invoked when a
3848 * completion event occurs on the CQ.
3849 * @event_handler: A user-specified callback that is invoked when an
3850 * asynchronous event not associated with a completion occurs on the CQ.
3851 * @cq_context: Context associated with the CQ returned to the user via
3852 * the associated completion and event handlers.
3853 * @cq_attr: The attributes the CQ should be created upon.
3855 * Users can examine the cq structure to determine the actual CQ size.
3857 struct ib_cq *__ib_create_cq(struct ib_device *device,
3858 ib_comp_handler comp_handler,
3859 void (*event_handler)(struct ib_event *, void *),
3861 const struct ib_cq_init_attr *cq_attr,
3862 const char *caller);
3863 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3864 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3867 * ib_resize_cq - Modifies the capacity of the CQ.
3868 * @cq: The CQ to resize.
3869 * @cqe: The minimum size of the CQ.
3871 * Users can examine the cq structure to determine the actual CQ size.
3873 int ib_resize_cq(struct ib_cq *cq, int cqe);
3876 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3877 * @cq: The CQ to modify.
3878 * @cq_count: number of CQEs that will trigger an event
3879 * @cq_period: max period of time in usec before triggering an event
3882 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3885 * ib_destroy_cq_user - Destroys the specified CQ.
3886 * @cq: The CQ to destroy.
3887 * @udata: Valid user data or NULL for kernel objects
3889 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3892 * ib_destroy_cq - Destroys the specified kernel CQ.
3893 * @cq: The CQ to destroy.
3895 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3897 static inline void ib_destroy_cq(struct ib_cq *cq)
3899 ib_destroy_cq_user(cq, NULL);
3903 * ib_poll_cq - poll a CQ for completion(s)
3904 * @cq:the CQ being polled
3905 * @num_entries:maximum number of completions to return
3906 * @wc:array of at least @num_entries &struct ib_wc where completions
3909 * Poll a CQ for (possibly multiple) completions. If the return value
3910 * is < 0, an error occurred. If the return value is >= 0, it is the
3911 * number of completions returned. If the return value is
3912 * non-negative and < num_entries, then the CQ was emptied.
3914 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3917 return cq->device->ops.poll_cq(cq, num_entries, wc);
3921 * ib_req_notify_cq - Request completion notification on a CQ.
3922 * @cq: The CQ to generate an event for.
3924 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3925 * to request an event on the next solicited event or next work
3926 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3927 * may also be |ed in to request a hint about missed events, as
3931 * < 0 means an error occurred while requesting notification
3932 * == 0 means notification was requested successfully, and if
3933 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3934 * were missed and it is safe to wait for another event. In
3935 * this case is it guaranteed that any work completions added
3936 * to the CQ since the last CQ poll will trigger a completion
3937 * notification event.
3938 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3939 * in. It means that the consumer must poll the CQ again to
3940 * make sure it is empty to avoid missing an event because of a
3941 * race between requesting notification and an entry being
3942 * added to the CQ. This return value means it is possible
3943 * (but not guaranteed) that a work completion has been added
3944 * to the CQ since the last poll without triggering a
3945 * completion notification event.
3947 static inline int ib_req_notify_cq(struct ib_cq *cq,
3948 enum ib_cq_notify_flags flags)
3950 return cq->device->ops.req_notify_cq(cq, flags);
3954 * ib_req_ncomp_notif - Request completion notification when there are
3955 * at least the specified number of unreaped completions on the CQ.
3956 * @cq: The CQ to generate an event for.
3957 * @wc_cnt: The number of unreaped completions that should be on the
3958 * CQ before an event is generated.
3960 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3962 return cq->device->ops.req_ncomp_notif ?
3963 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3968 * ib_dma_mapping_error - check a DMA addr for error
3969 * @dev: The device for which the dma_addr was created
3970 * @dma_addr: The DMA address to check
3972 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3974 return dma_mapping_error(dev->dma_device, dma_addr);
3978 * ib_dma_map_single - Map a kernel virtual address to DMA address
3979 * @dev: The device for which the dma_addr is to be created
3980 * @cpu_addr: The kernel virtual address
3981 * @size: The size of the region in bytes
3982 * @direction: The direction of the DMA
3984 static inline u64 ib_dma_map_single(struct ib_device *dev,
3985 void *cpu_addr, size_t size,
3986 enum dma_data_direction direction)
3988 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3992 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3993 * @dev: The device for which the DMA address was created
3994 * @addr: The DMA address
3995 * @size: The size of the region in bytes
3996 * @direction: The direction of the DMA
3998 static inline void ib_dma_unmap_single(struct ib_device *dev,
3999 u64 addr, size_t size,
4000 enum dma_data_direction direction)
4002 dma_unmap_single(dev->dma_device, addr, size, direction);
4006 * ib_dma_map_page - Map a physical page to DMA address
4007 * @dev: The device for which the dma_addr is to be created
4008 * @page: The page to be mapped
4009 * @offset: The offset within the page
4010 * @size: The size of the region in bytes
4011 * @direction: The direction of the DMA
4013 static inline u64 ib_dma_map_page(struct ib_device *dev,
4015 unsigned long offset,
4017 enum dma_data_direction direction)
4019 return dma_map_page(dev->dma_device, page, offset, size, direction);
4023 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4024 * @dev: The device for which the DMA address was created
4025 * @addr: The DMA address
4026 * @size: The size of the region in bytes
4027 * @direction: The direction of the DMA
4029 static inline void ib_dma_unmap_page(struct ib_device *dev,
4030 u64 addr, size_t size,
4031 enum dma_data_direction direction)
4033 dma_unmap_page(dev->dma_device, addr, size, direction);
4037 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4038 * @dev: The device for which the DMA addresses are to be created
4039 * @sg: The array of scatter/gather entries
4040 * @nents: The number of scatter/gather entries
4041 * @direction: The direction of the DMA
4043 static inline int ib_dma_map_sg(struct ib_device *dev,
4044 struct scatterlist *sg, int nents,
4045 enum dma_data_direction direction)
4047 return dma_map_sg(dev->dma_device, sg, nents, direction);
4051 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4052 * @dev: The device for which the DMA addresses were created
4053 * @sg: The array of scatter/gather entries
4054 * @nents: The number of scatter/gather entries
4055 * @direction: The direction of the DMA
4057 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4058 struct scatterlist *sg, int nents,
4059 enum dma_data_direction direction)
4061 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4064 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4065 struct scatterlist *sg, int nents,
4066 enum dma_data_direction direction,
4067 unsigned long dma_attrs)
4069 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4073 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4074 struct scatterlist *sg, int nents,
4075 enum dma_data_direction direction,
4076 unsigned long dma_attrs)
4078 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4082 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4083 * @dev: The device to query
4085 * The returned value represents a size in bytes.
4087 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4089 return dma_get_max_seg_size(dev->dma_device);
4093 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4094 * @dev: The device for which the DMA address was created
4095 * @addr: The DMA address
4096 * @size: The size of the region in bytes
4097 * @dir: The direction of the DMA
4099 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4102 enum dma_data_direction dir)
4104 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4108 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4109 * @dev: The device for which the DMA address was created
4110 * @addr: The DMA address
4111 * @size: The size of the region in bytes
4112 * @dir: The direction of the DMA
4114 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4117 enum dma_data_direction dir)
4119 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4123 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4124 * @dev: The device for which the DMA address is requested
4125 * @size: The size of the region to allocate in bytes
4126 * @dma_handle: A pointer for returning the DMA address of the region
4127 * @flag: memory allocator flags
4129 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4131 dma_addr_t *dma_handle,
4134 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4138 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4139 * @dev: The device for which the DMA addresses were allocated
4140 * @size: The size of the region
4141 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4142 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4144 static inline void ib_dma_free_coherent(struct ib_device *dev,
4145 size_t size, void *cpu_addr,
4146 dma_addr_t dma_handle)
4148 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4152 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4153 * HCA translation table.
4154 * @mr: The memory region to deregister.
4155 * @udata: Valid user data or NULL for kernel object
4157 * This function can fail, if the memory region has memory windows bound to it.
4159 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4162 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4163 * HCA translation table.
4164 * @mr: The memory region to deregister.
4166 * This function can fail, if the memory region has memory windows bound to it.
4168 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4170 static inline int ib_dereg_mr(struct ib_mr *mr)
4172 return ib_dereg_mr_user(mr, NULL);
4175 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4176 u32 max_num_sg, struct ib_udata *udata);
4178 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4179 enum ib_mr_type mr_type, u32 max_num_sg)
4181 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4184 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4185 u32 max_num_data_sg,
4186 u32 max_num_meta_sg);
4189 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4191 * @mr - struct ib_mr pointer to be updated.
4192 * @newkey - new key to be used.
4194 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4196 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4197 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4201 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4202 * for calculating a new rkey for type 2 memory windows.
4203 * @rkey - the rkey to increment.
4205 static inline u32 ib_inc_rkey(u32 rkey)
4207 const u32 mask = 0x000000ff;
4208 return ((rkey + 1) & mask) | (rkey & ~mask);
4212 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4213 * @pd: The protection domain associated with the unmapped region.
4214 * @mr_access_flags: Specifies the memory access rights.
4215 * @fmr_attr: Attributes of the unmapped region.
4217 * A fast memory region must be mapped before it can be used as part of
4220 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4221 int mr_access_flags,
4222 struct ib_fmr_attr *fmr_attr);
4225 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4226 * @fmr: The fast memory region to associate with the pages.
4227 * @page_list: An array of physical pages to map to the fast memory region.
4228 * @list_len: The number of pages in page_list.
4229 * @iova: The I/O virtual address to use with the mapped region.
4231 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4232 u64 *page_list, int list_len,
4235 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4239 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4240 * @fmr_list: A linked list of fast memory regions to unmap.
4242 int ib_unmap_fmr(struct list_head *fmr_list);
4245 * ib_dealloc_fmr - Deallocates a fast memory region.
4246 * @fmr: The fast memory region to deallocate.
4248 int ib_dealloc_fmr(struct ib_fmr *fmr);
4251 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4252 * @qp: QP to attach to the multicast group. The QP must be type
4254 * @gid: Multicast group GID.
4255 * @lid: Multicast group LID in host byte order.
4257 * In order to send and receive multicast packets, subnet
4258 * administration must have created the multicast group and configured
4259 * the fabric appropriately. The port associated with the specified
4260 * QP must also be a member of the multicast group.
4262 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4265 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4266 * @qp: QP to detach from the multicast group.
4267 * @gid: Multicast group GID.
4268 * @lid: Multicast group LID in host byte order.
4270 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4273 * ib_alloc_xrcd - Allocates an XRC domain.
4274 * @device: The device on which to allocate the XRC domain.
4275 * @caller: Module name for kernel consumers
4277 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4278 #define ib_alloc_xrcd(device) \
4279 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4282 * ib_dealloc_xrcd - Deallocates an XRC domain.
4283 * @xrcd: The XRC domain to deallocate.
4284 * @udata: Valid user data or NULL for kernel object
4286 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4288 static inline int ib_check_mr_access(int flags)
4291 * Local write permission is required if remote write or
4292 * remote atomic permission is also requested.
4294 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4295 !(flags & IB_ACCESS_LOCAL_WRITE))
4301 static inline bool ib_access_writable(int access_flags)
4304 * We have writable memory backing the MR if any of the following
4305 * access flags are set. "Local write" and "remote write" obviously
4306 * require write access. "Remote atomic" can do things like fetch and
4307 * add, which will modify memory, and "MW bind" can change permissions
4308 * by binding a window.
4310 return access_flags &
4311 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4312 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4316 * ib_check_mr_status: lightweight check of MR status.
4317 * This routine may provide status checks on a selected
4318 * ib_mr. first use is for signature status check.
4320 * @mr: A memory region.
4321 * @check_mask: Bitmask of which checks to perform from
4322 * ib_mr_status_check enumeration.
4323 * @mr_status: The container of relevant status checks.
4324 * failed checks will be indicated in the status bitmask
4325 * and the relevant info shall be in the error item.
4327 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4328 struct ib_mr_status *mr_status);
4331 * ib_device_try_get: Hold a registration lock
4332 * device: The device to lock
4334 * A device under an active registration lock cannot become unregistered. It
4335 * is only possible to obtain a registration lock on a device that is fully
4336 * registered, otherwise this function returns false.
4338 * The registration lock is only necessary for actions which require the
4339 * device to still be registered. Uses that only require the device pointer to
4340 * be valid should use get_device(&ibdev->dev) to hold the memory.
4343 static inline bool ib_device_try_get(struct ib_device *dev)
4345 return refcount_inc_not_zero(&dev->refcount);
4348 void ib_device_put(struct ib_device *device);
4349 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4350 enum rdma_driver_id driver_id);
4351 struct ib_device *ib_device_get_by_name(const char *name,
4352 enum rdma_driver_id driver_id);
4353 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4354 u16 pkey, const union ib_gid *gid,
4355 const struct sockaddr *addr);
4356 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4358 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4360 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4361 struct ib_wq_init_attr *init_attr);
4362 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4363 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4365 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4366 struct ib_rwq_ind_table_init_attr*
4367 wq_ind_table_init_attr);
4368 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4370 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4371 unsigned int *sg_offset, unsigned int page_size);
4372 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4373 int data_sg_nents, unsigned int *data_sg_offset,
4374 struct scatterlist *meta_sg, int meta_sg_nents,
4375 unsigned int *meta_sg_offset, unsigned int page_size);
4378 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4379 unsigned int *sg_offset, unsigned int page_size)
4383 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4389 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4390 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4392 void ib_drain_rq(struct ib_qp *qp);
4393 void ib_drain_sq(struct ib_qp *qp);
4394 void ib_drain_qp(struct ib_qp *qp);
4396 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4398 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4400 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4401 return attr->roce.dmac;
4405 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4407 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4408 attr->ib.dlid = (u16)dlid;
4409 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4410 attr->opa.dlid = dlid;
4413 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4415 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4416 return attr->ib.dlid;
4417 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4418 return attr->opa.dlid;
4422 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4427 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4432 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4435 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4436 attr->ib.src_path_bits = src_path_bits;
4437 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4438 attr->opa.src_path_bits = src_path_bits;
4441 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4443 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4444 return attr->ib.src_path_bits;
4445 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4446 return attr->opa.src_path_bits;
4450 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4453 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4454 attr->opa.make_grd = make_grd;
4457 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4459 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4460 return attr->opa.make_grd;
4464 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4466 attr->port_num = port_num;
4469 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4471 return attr->port_num;
4474 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4477 attr->static_rate = static_rate;
4480 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4482 return attr->static_rate;
4485 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4486 enum ib_ah_flags flag)
4488 attr->ah_flags = flag;
4491 static inline enum ib_ah_flags
4492 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4494 return attr->ah_flags;
4497 static inline const struct ib_global_route
4498 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4503 /*To retrieve and modify the grh */
4504 static inline struct ib_global_route
4505 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4510 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4512 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4514 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4517 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4520 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4522 grh->dgid.global.subnet_prefix = prefix;
4525 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4528 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4530 grh->dgid.global.interface_id = if_id;
4533 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4534 union ib_gid *dgid, u32 flow_label,
4535 u8 sgid_index, u8 hop_limit,
4538 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4540 attr->ah_flags = IB_AH_GRH;
4543 grh->flow_label = flow_label;
4544 grh->sgid_index = sgid_index;
4545 grh->hop_limit = hop_limit;
4546 grh->traffic_class = traffic_class;
4547 grh->sgid_attr = NULL;
4550 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4551 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4552 u32 flow_label, u8 hop_limit, u8 traffic_class,
4553 const struct ib_gid_attr *sgid_attr);
4554 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4555 const struct rdma_ah_attr *src);
4556 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4557 const struct rdma_ah_attr *new);
4558 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4561 * rdma_ah_find_type - Return address handle type.
4563 * @dev: Device to be checked
4564 * @port_num: Port number
4566 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4569 if (rdma_protocol_roce(dev, port_num))
4570 return RDMA_AH_ATTR_TYPE_ROCE;
4571 if (rdma_protocol_ib(dev, port_num)) {
4572 if (rdma_cap_opa_ah(dev, port_num))
4573 return RDMA_AH_ATTR_TYPE_OPA;
4574 return RDMA_AH_ATTR_TYPE_IB;
4577 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4581 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4582 * In the current implementation the only way to get
4583 * get the 32bit lid is from other sources for OPA.
4584 * For IB, lids will always be 16bits so cast the
4585 * value accordingly.
4589 static inline u16 ib_lid_cpu16(u32 lid)
4591 WARN_ON_ONCE(lid & 0xFFFF0000);
4596 * ib_lid_be16 - Return lid in 16bit BE encoding.
4600 static inline __be16 ib_lid_be16(u32 lid)
4602 WARN_ON_ONCE(lid & 0xFFFF0000);
4603 return cpu_to_be16((u16)lid);
4607 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4609 * @device: the rdma device
4610 * @comp_vector: index of completion vector
4612 * Returns NULL on failure, otherwise a corresponding cpu map of the
4613 * completion vector (returns all-cpus map if the device driver doesn't
4614 * implement get_vector_affinity).
4616 static inline const struct cpumask *
4617 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4619 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4620 !device->ops.get_vector_affinity)
4623 return device->ops.get_vector_affinity(device, comp_vector);
4628 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4629 * and add their gids, as needed, to the relevant RoCE devices.
4631 * @device: the rdma device
4633 void rdma_roce_rescan_device(struct ib_device *ibdev);
4635 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4637 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4639 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4640 enum rdma_netdev_t type, const char *name,
4641 unsigned char name_assign_type,
4642 void (*setup)(struct net_device *));
4644 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4645 enum rdma_netdev_t type, const char *name,
4646 unsigned char name_assign_type,
4647 void (*setup)(struct net_device *),
4648 struct net_device *netdev);
4651 * rdma_set_device_sysfs_group - Set device attributes group to have
4652 * driver specific sysfs entries at
4653 * for infiniband class.
4655 * @device: device pointer for which attributes to be created
4656 * @group: Pointer to group which should be added when device
4657 * is registered with sysfs.
4658 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4659 * group per device to have sysfs attributes.
4661 * NOTE: New drivers should not make use of this API; instead new device
4662 * parameter should be exposed via netlink command. This API and mechanism
4663 * exist only for existing drivers.
4666 rdma_set_device_sysfs_group(struct ib_device *dev,
4667 const struct attribute_group *group)
4669 dev->groups[1] = group;
4673 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4675 * @device: device pointer for which ib_device pointer to retrieve
4677 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4680 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4682 struct ib_core_device *coredev =
4683 container_of(device, struct ib_core_device, dev);
4685 return coredev->owner;
4689 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4690 * ib_device holder structure from device pointer.
4692 * NOTE: New drivers should not make use of this API; This API is only for
4693 * existing drivers who have exposed sysfs entries using
4694 * rdma_set_device_sysfs_group().
4696 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4697 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4699 bool rdma_dev_access_netns(const struct ib_device *device,
4700 const struct net *net);
4701 #endif /* IB_VERBS_H */